Análise de imagens de satélite para detecção de lixo: desafios e potencialidades para a gestão ambiental sustentável

Abstract

Two images from the CBERS 4A satellite, captured by the Wide Field Multispectral and Panchromatic Camera (WPM), with radiometric resolution of 8 bits, panchromatic spatial resolution of 2m, and multispectral spatial resolution of 8m of Edeia/GO and Indiara/GO municipalities, were obtained from the National Institute for Space Research (INPE) in April 2024 for detecting areas with irregular waste disposal, employing digital processing techniques in the QGIS software and statistical analyses. The main objective was to identify and characterize patterns of waste on the ground through spectral analysis of the blue, green, red, and near-infrared bands. Initial analysis revealed statistically significant differences between waste pixels and other landscape elements such as vegetation, roads, bare soil, residences, and water. Results from variance and T-test indicated that waste pixels exhibit distinct spectral characteristics compared to vegetation and water, but there is a moderate risk of confusion with roads and bare soil. Specifically, waste pixels were more distinguishable in the near-infrared and red bands, while showing greater similarity with roads and residences in the blue and green bands. To evaluate the waste detection model's performance, the Fβ score, which considers both precision and recall, was used. Results showed overall low precision due to challenges in distinguishing small waste deposits and limitations in image resolution. The analysis also highlighted that accurate classification of waste in the landscape is challenging due to spectral similarity with other objects and the presence of atmospheric noise and sensor limitations. The research demonstrated the feasibility of using artificial intelligence and deep learning to monitor areas with irregular waste disposal, but significant adjustments are needed to improve accuracy and reduce false positives and negatives. Expanding the dataset and refining models are recommended to enable more precise and reliable waste detection in the landscape, thereby contributing to the development of sustainable environmental management practices.