La classification d'une image ETM+ de Landsat a été réalisée pour la cartographie des types de couverture du sol dans la moitié sud du parc national du Canada Auyuittuq. Le projet fait appel à l'analyse d'image par objets spatiaux (object-based image analysis ). Le logiciel eCognition 4.0 permet une segmentation hiérarchique de l'image qui est analogue au concept de l'inventaire écologique des parcs nationaux. Un territoire d'environ 8 300 km 2 a été cartographié à trois niveaux de perception différents à partir d'une image acquise le 13 août 2000. Un modèle numérique d'altitude fut incorporé au projet et de nombreux indices spectraux ont été calculés à partir des données ETM + . Le niveau de segmentation brute comporte 375 312 objets regroupés en 36 classes. À ce niveau, la structure de classification repose sur 118 règles référant aux paramètres spectraux, spatiaux et topographiques des segments. Ces règles combinent des systèmes de seuillages chiffrés et des opérations de tri au plus proche voisin. L'attribution des segments aux classes du projet est tributaire de ces règles et répond à une logique floue. À la suite d'une fusion de segments et d'un premier regroupement de classes, on obtient le second niveau du projet, qui compte 102 239 objets et 28 classes thématiques. Ce niveau s'apparente aux"écotypes" au sens de l'inventaire écologique des parcs nationaux. Un second regroupement réduit à 9 le nombre de classes et à 36 887 le nombre d'objets, ce qui se rapproche d'une cartographie des «écosystèmes » de l'inventaire écologique. Sur le terrain, 315 relevés photographiques de la végétation ont été réalisés dans les vallées Akshayuk et Naqsaq. Pour chaque relevé, les pourcentages de couverture de 5 strates végétales ont été estimés, de manière à ranger les relevés dans 8 classes de végétation connues a priori. Dans l'image, ce sont 135 segments qui ont pu être retenus comme échantillons. De ce nombre, 71 et 64 échantillons furent retenus respectivement pour l'entraînement et la validation de la classification au plus proche voisin qui fut réalisée pour la végétation. L'exactitude générale de la classification de la végétation a été estimée à 54,7 %. Contrairement à la végétation, le couvert non-végétal est classifié suivant principalement un système de règles, lesquelles décrivent le comportement spectral de 34 types de surfaces selon une structure de classification hiérarchique.La classification des 20 écotypes non-végétaux a été validée par photo-interprétation à l'aide de 992 segments-non-végétale est évaluée à 83,2 %. Une fois synthétisée au niveau des écosystèmes, la classification atteint un taux de succès global de 92,7 %. Pour la classification de la végétation, l'analyse d'image par objets spatiaux livre une cartographie dont l'exactitude est équivalente à celle d'une classification basée sur le pixel réalisée par Parcs Canada pour la même image (54,7 % vs 53,4 %). Notre stratification comporte cependant un plus grand nombre de catégories non-végétales et leur classification atteint un niveau d'exactitude supérieur. L'analyse par objet spatiaux nous a permis d'aller au-delà de l'analyse pûrement spectrale pour incorporer des paramètres texturaux, géométriques et contextuels à la procédure de classification. Elle résulte en une représentation plus synthétique de l'information cartographique que la classification basée sur le pixel, mais les patrons spatiaux les plus fins des milieux les plus hétérogènes sont alors perdus.La structure de classification développée pour notre image peut être transposée avec succès vers une nouvelle image, mais ceci exige que soient apportés des ajustements aux règles de classification, voire l'ajout ou la suppréssion de certaines règles.La segmentation hiérarchique s'avère utile comme analogue au concept de l'inventaire écologique des parcs nationaux. Les informations véhiculées par chacun des niveaux de notre classification sont des intrants importants pour l'inventaire écologique du parc national du Canada Auyuittuq. Une typologie définitive reste à définir tant pour la classification de la végétation que pour le couvert non-végétal des parcs nationaux de l'Arctique. Des clés de classification seraient requises pour traduire ces typologies en paramètres reconnaissables sur le terrain. En mettant en commun les diverses données de terrain existantes pour le parc national du Canada Auyuittuq et en les structurant selon ces typologies, on obtiendrait une banque d'échantillons augmentée et plus cohérente. De telles données de références s'avéreraient une base solide pour la validation des classifications présentement disponibles ainsi que pour la mise en oeuvre de travaux futurs en matière d'inventaire écologique pour le parc national du Canada Auyuittuq.

Classification of airborne laser scanning data using information about intensity and width of the recorded signal Abstract One of the basic tasks in analysing airborne laser scanning (ALS) data is filtration of mass 3D point cloud with purpose to create digital terrain model and digital surface model. New scanner generation (so called Full-waveform LiDAR) allows analysing the whole recorded signal. The recorded value of amplitude and signal width accordant with reflectance of different objects differs according to geometry of the objects. Objective of this thesis is to create a methodology for classification of ALS data in settled areas. This methodology will be based on number of reflections, amplitude of reflected signal, recorded signal width and on spatial attributes. At the same time it will be analysed how the parameters of amplitude and signal width are affected by characteristics of estate surface. It means which radiometrical characteristics (e.g. different roof materials) and geometrical characteristics (e.g. different roof inclination) belong to which amplitude and signal width. Basic question of this thesis is if amplitude and signal width are good attributes to improve the quality of filtration of mass 3D point cloud in chosen area and if so, how. Key words: classification, segmentation, LiDAR,...

This master thesis deals with a classification technique for an automatic detection of different land cover types from combination of high resolution imagery and LiDAR data sets. The main aim is to introduce additional post-processing method to commonly accessible quality data sets which can replace traditional mapping techniques for certain type of applications. Classification is the process of dividing the image into land cover categories which helps with continuous and up-to-date monitoring management. Nowadays, with all the technologies and software available, it is possible to replace traditional monitoring methods with more automated processes to generate accurate and cost-effective results. This project uses object-oriented image analysis (OBIA) to classify available data sets into five main land cover classes. The automate classification rule set providing overall accuracy of 88% of correctly classified land cover types was developed and evaluated in this research. Further, the transferability of developed approach was tested upon the same type of data sets within different study area with similar success – overall accuracy was 87%. Also the limitations found during the investigation procedure are discussed and brief further approach in this field is outlined.

No description available.

No description available.

No description available.

With the advent of very high spatial resolution (VHR) satellite, spatial details within the image scene have increased considerably. This led to the development of object-based image analysis (OBIA) for the analysis of VHR satellite images. Image segmentation is the fundamental step for OBIA. However, a large number of techniques exist for RS image segmentation. To identify the best ones for VHR imagery, a comprehensive literature review on image segmentation is performed. Based on that review, it is found that the multiresolution segmentation, as implemented in the commercial software eCognition, is the most widely-used technique and has been successfully applied for wide variety of VHR images. However, the multiresolution segmentation suffers from the parameter estimation problem. Therefore, this study proposes a solution to the problem of the parameter estimation for improving its efficiency in VHR image segmentation. The solution aims to identify the optimal parameters, which correspond to optimal segmentation. The solution to the parameter estimation is drawn from the Equations related to the merging of any two adjacent objects in multiresolution segmentation. The solution utilizes spectral, shape, size, and neighbourhood relationships for a supervised solution. In order to justify the results of the solution, a global segmentation accuracy evaluation technique is also proposed. The solution performs excellently with the VHR images of different sensors, scenes, and land cover classes. In order to justify the applicability of solution to a real life problem, a building detection application based on multiresolution segmentation from the estimated parameters, is carried out. The accuracy of the building detection is found nearly to be eighty percent. Finally, it can be concluded that the proposed solution is fast, easy to implement and effective for the intended applications.

Orientador: José Raimundo de Oliveira / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-19T12:19:26Z (GMT). No. of bitstreams: 1 Mamani_AnaBeatrizAlvarez_M.pdf: 4159333 bytes, checksum: 1e7bbac608fe9a8dc553adedc4b721b7 (MD5) Previous issue date: 2011 / Resumo: Esta tese apresenta uma nova e efetiva abordagem neuro-genética denominada MAAM-GA constituída por um algoritmo genético e uma rede neural associativa morfológica para a solução de problemas de reconhecimento de padrões. Especificamente, uma rede neural associativa morfológica é combinada com um algoritmo genético que é utilizado na construção da rede neural com a finalidade de aumentar a eficiência e robustez no reconhecimento de padrões. Um estudo detalhado do desempenho da abordagem é apresentado, utilizando imagens em níveis de cinza como padrões. Resultados numéricos e visuais da recuperação dos padrões são apresentados e o desempenho alcançado é comparado com outros modelos neurais associativos morfológicos relevantes para padrões de valor real, mostrando a eficiência e a robustez da abordagem proposta na recordação de imagens em níveis de cinza. Esta abordagem faz parte do desenvolvimento dos sistemas inteligentes que impulsionam o avanço de outras áreas. Pensando em uma potencial aplicação, a proposta neuro-genética é utilizada para resolver o problema de reconhecimento de gestos da mão. O reconhecimento de gestos é um caminho natural de interação humano-computador, e considerando a diversidade e a diferença manifestada pelo ser humano, para muitas pessoas que possuem deficiência física e sensorial, os gestos da mão são o meio principal de comunicação. Várias tecnologias têm sido propostas para trazer benefícios às pessoas com limitações de comunicação. Os ambientes inteligentes surgiram com o principal propósito de melhorar a qualidade de vida do ser humano baseados em ferramentas computacionais, facilitando o desenvolvimento de processos e ações de nosso cotidiano. O reconhecimento de gestos da mão é uma função do ambiente inteligente. Assim, para pessoas portadoras de deficiências físicas que limitem a sua comunicação oral, o reconhecimento de gestos em um ambiente inteligente poderá lhes trazer múltiplos benefícios na comunicação, interação e acessibilidade, permitindo a sua integração com o ambiente. Embora preocupados com pessoas portadoras de deficiências físicas, o sistema de reconhecimento de gestos da mão como parte de um ambiente inteligente destina-se, sobretudo a beneficiar todo e qualquer cidadão que dele tenha acesso. Assim, nesta tese é apresentado um estudo de um sistema de reconhecimento de gestos da mão baseado em visão artificial capaz de reconhecer gestos estáticos específicos da mão. Este sistema foi dividido em três módulos, módulo de detecção e segmentação, módulo de extração de características e o módulo de identificação e reconhecimento propriamente dito que utiliza a abordagem neuro-genética proposta. Métodos utilizados no pré-processamento das imagens para segmentação e caracterização também são apresentados. Resultados alcançados com a abordagem proposta são muito incentivadores e sugerem que a proposta possa ser considerada como uma ferramenta eficiente e robusta para recuperação e identificação a ser usada em diversas aplicações relacionadas à interface natural humano-computador. O ótimo desempenho do sistema é um passo para continuar na busca de novas tecnologias para criar um ambiente inteligente que dê suporte às necessidades de pessoas com deficiência visual, auditiva ou motora lhes dando certo nível de autonomia, capacidade de controle do entorno e de comunicação / Abstract: This thesis presents an innovative approach to solving problems of pattern recognition using a neural-genetic combination. Specifically, a morphological associative neural network is combined with a genetic algorithm that is used in the construction of the neural network for increasing the efficiency and robustness of pattern recall. A detailed study about the performance of the approach is presented, using grayscale images as patterns. Numerical and visual results are presented and the performance achieved is compared with other morphological associative neural models showing its effectiveness and robustness in the grayscale images recall. Thinking about a potential application, the proposed approach is used to solve the problem of hand gestures recognition. The hand gestures recognition is a natural way of human-computer interaction and considering the diversity and difference manifested by the human, for many people who have physical and sensory disabilities, the hand gestures is the primary means of communication. Several technologies have been proposed to bring benefits to people with limited communication. The intelligent environments emerged with the main purpose of improving the quality of human life based in computational tools facilitating the development of processes and actions of everyday life. The hand gestures recognition is a function of intelligent environments. So, for people with physical disabilities that limit their oral communication gesture recognition in an intelligent environment can take many benefits in communication, interaction and accessibility allowing its integration with the environment. Although concerned about people with disabilities, the hand gestures recognition system is mainly intended to benefit every people who has access to the environment. Thus, this thesis presents a study of a hand gestures recognition system. The system is able to recognize static hand gestures using the proposed Neural-Genetic Approach. Methods used in the image preprocessing and characterization are also presented. Results achieved with the proposed approach are very encouraging and suggest that the proposal can be considered as an efficient and robust tool for recovery and identification to be used in various applications related to natural human-computer interface. The optimal system performance is a big step to continue the search for new technologies to create an intelligent environment that supports the needs of people with visual, hearing or motor disability / Mestrado / Engenharia de Computação / Doutor em Engenharia Elétrica

In a world with an ever growing population, the ability to grow food eciently is essential. One way to improve the eciency is by automation. The purpose of this project is therefore to investigate how the identification of a plant’s stage in its growth cycle that can be made autonomous. This was done with the method of measuring the amount of green pixels in an image of the plant. To be able to answer our research questions a demonstrator was built. The demonstrator is a greenhouse with a non regulated aeroponic system, a regulation system for humidity and an identification system for determining the plant growth stage. The plant chosen to test the identification system was basil. The identification system successfully identified the stage of plants well into the adult stage, in the seed stage and in the middle of the sprout stage. It was however not always successful in the identification of plants transitioning from the sprout stage into the adult stage. / I en värld med en ständigt växande befolkning är förmågan att odla mat effektivt nödvändig. En metod för att öka denna effektivitet är genom automatisering. Syftet för detta projekt är därför att undersöka hur identifieringen av en plantans stadie i dess växtcykel kan automatiseras. Detta gjordes genom att mäta antalet gröna pixlar i en bild av plantan. För att kunna svara våra forskningsfrågor byggdes en testmiljö. Testmiljön bestod av ett växthus med ett oreglerat aeroponiskt system, ett regulationssystem för luftfuktighet och ett identifikationsssystem för att avgöra en plantas stadie i dess växtcykel. Plantan som valdes för att testa identifikationssystemet var basilika. Identifikationssystemet som togs fram kunde med framgång identifiera stadiet av en planta som är långt in i dess vuxna stadie, i förstadiet eller i mitten av dess groddstadie. Plantor som precis övergått från grodd till vuxet stadie blev däremot inte alltid identifierade korrekt.

Expansion of Pinus L. (pinyon) and Juniperus L. (juniper) (P-J) trees into sagebrush (Artemisia L.) steppe communities can lead to negative effects on hydrology, loss of wildlife habitat, and a decrease in desirable understory vegetation. Tree reduction treatments are often implemented to mitigate these negative effects. In order to prioritize and effectively plan these treatments, rapid, accurate, and inexpensive methods are needed to estimate tree canopy cover at the landscape scale. We used object based image analysis (OBIA) software (Feature AnalystTM for ArcMap 10.1®, ENVI Feature Extraction®, and Trimble eCognition Developer 8.2®) to extract tree canopy cover using NAIP (National Agricultural Imagery Program) imagery. We then compared our extractions with ground measured tree canopy cover (crown diameter and line point) on 309 subplots across 44 sites in Utah. Extraction methods did not consistently over- or under-estimate ground measured P-J canopy cover except where tree cover was > 45%. Estimates of tree canopy cover using OBIA techniques were strongly correlated with estimates using the crown diameter method (r = 0.93 for ENVI, 0.91 for Feature Analyst, and 0.92 for eCognition). Tree cover estimates using OBIA techniques had lower correlations with tree cover measurements using the line-point method (r = 0.85 for ENVI, 0.83 for Feature Analyst, and 0.83 for eCognition). Results from this study suggest that OBIA techniques may be used to extract P-J tree canopy cover accurately and inexpensively. All software packages accurately evaluated accurately extracted P-J canopy cover from NAIP imagery when imagery was not blurred and when P-J cover was not mixed with Amelanchier alnifolia (Utah serviceberry) and Quercus gambelii (Gambel's oak), which are shrubs with similar spectral values as P-J.