Search

Multiresolution deep learning approaches, such as the U-Net architecture, have achieved high performance in classifying and segmenting images. However, these approaches do not provide a latent image representation and cannot be used to decompose, denoise, and reconstruct image data. The U-Net and other convolutional neural network (CNNs) architectures commonly use pooling to enlarge the receptive field, which usually results in irreversible information loss. This study proposes to include a Riesz-Quincunx (RQ) wavelet transform, which combines 1) higher-order Riesz wavelet transform and 2) orthogonal Quincunx wavelets (which have both been used to reduce blur in medical images) inside the U-net architecture, to reduce noise in satellite images and their time-series. In the transformed feature space, we propose a variational approach to understand how random perturbations of the features affect the image to further reduce noise. Combining both approaches, we introduce a hybrid RQUNet-VAE scheme for image and time series decomposition used to reduce noise in satellite imagery. We present qualitative and quantitative experimental results that demonstrate that our proposed RQUNet-VAE was more effective at reducing noise in satellite imagery compared to other state-of-the-art methods. We also apply our scheme to several applications for multi-band satellite images, including: image denoising, image and time-series decomposition by diffusion and image segmentation., Comment: Submitted to IEEE Transactions on Geoscience and Remote Sensing (TGRS)

Land degradation is one of the most serious environmental problems of our time. Land degradation describes circumstances of reduced biological productivity. The fundamental goal of this thesis was to develop land degradation monitoring approaches based on remotely sensed estimates of vegetation production, which are capable of distinguishing human impacts from the effects of natural climatic and spatial variability. Communal homelands in South Africa (SA) are widely regarded to be severely degraded and the existence adjacent, non-degraded areas with the same soils and climate, provides a unique opportunity to test regional land degradation monitoring methods. The relationship between 1km AVHRR, growth season sumNDVI and herbaceous biomass measurements (1989-2003) was firstly tested in Kruger National Park, SA. The relationship was moderately strong, but weaker than expected. This was attributed to the fact that the small areas sampled at field sites were not representative of the spatial variability within 1x1km. The sumNDVI adequately estimated inter-annual changes in vegetation production and should therefore be useful for monitoring land degradation. Degraded areas mapped by the National-Land-Cover in north-eastern SA were compared to non-degraded areas in the same land capability units. The sumNDVI of the degraded areas was consistently lower, regardless of large variations in rainfall. However, the ecological stability and resilience of the degraded areas, as measured by the annual deviations from each pixel's mean sumNDVI, were no different to those of non-degraded areas. This suggests that the degraded areas may be in an alternative, but stable ecological state. To monitor human-induced land degradation it is essential to control for the effects of rainfall on vegetation production. Two methods were tested (i) Rain-Use Efficiency (RUE=NPP/Rainfall) and (ii) negative trends in the differences between the observed sumNDVI and the sumNDVI predicted by the rainfa

The inundated area of the Okavango Delta changes annually and interannually. The variability relates to regional precipitation over the catchment area in the Angolan highlands, and to local rainfall. The patterns of the wetland were captured using more than 3000 satellite images for the period 1972 to 2000, near daily NOAA AVHRR data for 1985-2000, and less frequent images of the Landsat sensors from 1972 onwards. One AVHRR image for every 10-day period was classified into land and water using an unsupervised classification method. Evaluation against Landsat TM and ERS2-ATSR data indicate an agreement of 89% for the size of estimated inundation area. Results show that the wetland area has varied between approximately 2450 km(2) and 11 400 km(2) during the last 30 years.