In this study, we used images obtained by Unmanned Aerial Vehicles (UAV) and an instance segmentation model based on deep learning (Mask R-CNN) to evaluate the ability to detect and delineate canopies in high density orange plantations. The main objective of the work was to evaluate the improvement acquired by the segmentation model when integrating the Canopy Height Model (CHM) as a fourth band to the images. Two models were evaluated, one with RGB images and the other with RGB + CHM images, and the results indicated that the model with combined images presents better results (overall accuracy from 90.42% to 97.01%). In addition to the comparison, this work suggests a more efficient ground truth mapping method and proposes a methodology for mosaicking the results by Mask R-CNN on remotely sensed images.

Landslides are recurrent events in Brazil, usually triggered by intense rainfall. When these events occur in urban areas, they end up becoming disasters due to economic damage, social impact, and loss of human life. The identification and monitoring of landslide-prone areas are crucial to avoid fatalities. Therefore, the aims of this work are a temporal analysis of the Factor of Safety variation in Campos do Jordão, using the mathematical model TRIGRS. During the analyzed period, two heavy rainfall events were recorded in the area and triggered landslides. The results show TRIGRS efficiency in correctly identify landslide-prone areas and its applicability to become a useful tool for urban planning and early warning systems.

As técnicas de análise de imagens baseadas em objetos e os sensores de alta resolução espacial têm sido muito usadas para mapear alvos urbanos. Uma das principais limitações de tais sensores tem sido a baixa resolução espectral, o que dificulta a discriminação de objetos urbanos com respostas espectrais semelhantes. Neste caso, dados auxiliares, tais como os LIDAR, têm sido usados para melhorar os resultados da classificação. O sensor WorldView-II, por possuir oito bandas espectrais, pode ser uma boa solução para a discriminação de alvos urbanos de difícil identificação somente com o uso de imagens de sensores de alta resolução com menor resolução espectral. Por outro lado, o sistema de interpretação de imagens InterIMAGE, que tem sido desenvolvido pela PUC-RJ em cooperação com o INPE, apresenta grande potencial para a classificação de áreas urbanas complexas. Neste contexto, este trabalho propõe avaliar o potencial das imagens do sensor WorldView-II para o mapeamento da cobertura do solo em uma área urbana do Trecho Oeste do Rodoanel Mário Covas, na Região Metropolitana de São Paulo, usando o InterIMAGE. O modelo de classificação é construído de acordo com a estratégia de análise de imagens do InterIMAGE. Os mapas temáticos de cobertura do solo obtidos apresentaram altos valores de exatidão global e índice Kappa iguais a 0,87 e 0,85, respectivamente. Além disso, alguns conflitos típicos da classificação de alvos urbanos foram resolvidos com a identificação de quinze classes de cobertura do solo. De uma forma geral, os resultados mostraram que as novas bandas espectrais do sensor WorldView-II foram fundamentais para a discriminação de alguns alvos urbanos, tais como Telha Cerâmica e Solo Exposto, geralmente, difíceis de serem identificados com os sensores de alta resolução espacial como o QuickBird-II e IKONOS.

The objective of this study is to estimate the canopy height of three sugarcane varieties at different growth stages, with UAV data and to evaluate its relationship with two vegetation indices (VIs) (NDVI and EVI) at different spatial resolutions (3m, 10m and 30m). The indices were calculated using images from the PlanetScope, Sentinel-2, and Landsat 8 satellites, acquired as close as possible to the UAV imaging date. The estimated canopy height for each field was obtained by subtracting the Digital Surface Model (DSM) from the Digital Terrain Model (DTM), built by the Structure from Motion (SfM) technique with UAV RGB images as input. The average from each estimated height was compared with the average measured in the field, to verify the accuracy of the model. Both Pearson’s correlation and the Determination Coefficient (R²) were calculated between the estimated heights and the VIs. The average estimated canopy height and measurements in the field were different (p<0.05), with the model generally underestimating the height. However, the plantation’s surface models portrayed the spatial variability within the field. The use of GCPs is mandatory to reduce errors in estimation. Regarding the indices, the spatial resolution did not influence the correlation analysis, with NDVI showing higher values than EVI, except for area A. However, all values, for both coefficients, were below 0.5 for all areas. Despite that, a temporal analysis is necessary to improve the relationship between the canopy height and VIs. The potential of UAV data as a proxy to zonal management should be addressed in future studies.