Publicação
CLASSIFICAÇÃO DE IMAGENS DE MUITO ALTA RESOLUÇÃO ESPACIAL PARA MAPEAMENTO DA ESPÉCIE INVASORA ACACIA DEALBATA
| Resumo: | In Portugal, the increasing proliferation of invasive alien species, such as Acacia dealbata, threatens the integrity of ecosystems, compromising native flora and fauna, and is considered one of the most problematic invasive species in the country.Remote sensing constitutes an essential tool for mapping invasive alien species, providing fundamental support for strategies of control and impact management. In addition to enabling large-scale monitoring, it represents an effective and lower-cost approach when compared to fieldwork, capable of covering vast areas with a high level of detail, particularly when very high-resolution imagery is used.The present study aimed to develop and validate a methodology for mapping the spatial distribution of the invasive alien species Acacia dealbata, using very high-resolution imagery from the Pléiades Neo satellite (0.3 m). The approach relied on pixel-based supervised classification, employing the Random Forest algorithm, based on spectral characteristics for identification and differentiation.Three experiments were conducted:• Experiment 1 was applied to the Lousã region, using approximately 4 million pixels as training data. The classifier achieved an overall accuracy of 89%.• Experiment 2, also applied to the Lousã region, aimed to increase the amount of training data in order to verify whether classification results would improve. However, despite the larger training dataset of around 9 million pixels, this experiment showed reduced performance, with an overall accuracy of 86%. This result demonstrates that the quality and representativeness of training data may be more decisive than the dataset size.• Experiment 3 was designed to assess the model’s generalization capacity, using a classification area different from the one where the training data were created. For this purpose, the model from Experiment 1 was applied, as it had shown the best results, to a new study area in the city of Coimbra. The model achieved an overall accuracy of 86%.The results demonstrate the capability of the proposed methodology to identify and delineate areas invaded by Acacia dealbata with high accuracy, supporting management, control, and prevention actions. This study reinforces the potential of remote sensing as a solution to support ecosystem conservation. |
|---|---|
| Autores principais: | Santos, Rafael Pereira |
| Assunto: | Remote sensing Acacia dealbata Pléiades Neo Supervised classification Random Forest Deteção remota Acacia dealbata Pléiades Neo Classificação supervisionada Random Forest |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Coimbra |
| Idioma: | português |
| Origem: | Estudo Geral - Universidade de Coimbra |
| Resumo: | In Portugal, the increasing proliferation of invasive alien species, such as Acacia dealbata, threatens the integrity of ecosystems, compromising native flora and fauna, and is considered one of the most problematic invasive species in the country.Remote sensing constitutes an essential tool for mapping invasive alien species, providing fundamental support for strategies of control and impact management. In addition to enabling large-scale monitoring, it represents an effective and lower-cost approach when compared to fieldwork, capable of covering vast areas with a high level of detail, particularly when very high-resolution imagery is used.The present study aimed to develop and validate a methodology for mapping the spatial distribution of the invasive alien species Acacia dealbata, using very high-resolution imagery from the Pléiades Neo satellite (0.3 m). The approach relied on pixel-based supervised classification, employing the Random Forest algorithm, based on spectral characteristics for identification and differentiation.Three experiments were conducted:• Experiment 1 was applied to the Lousã region, using approximately 4 million pixels as training data. The classifier achieved an overall accuracy of 89%.• Experiment 2, also applied to the Lousã region, aimed to increase the amount of training data in order to verify whether classification results would improve. However, despite the larger training dataset of around 9 million pixels, this experiment showed reduced performance, with an overall accuracy of 86%. This result demonstrates that the quality and representativeness of training data may be more decisive than the dataset size.• Experiment 3 was designed to assess the model’s generalization capacity, using a classification area different from the one where the training data were created. For this purpose, the model from Experiment 1 was applied, as it had shown the best results, to a new study area in the city of Coimbra. The model achieved an overall accuracy of 86%.The results demonstrate the capability of the proposed methodology to identify and delineate areas invaded by Acacia dealbata with high accuracy, supporting management, control, and prevention actions. This study reinforces the potential of remote sensing as a solution to support ecosystem conservation. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.