Abstract: Training an estimation model using soft labels includes receiving an image. It further includes generating a continuous target map corresponding to the image that includes hard labels and soft labels. A model is trained using the corresponding continuous target map.

Abstract: A set of images is received. A thermal anomaly is detected in the received set of images. The detected thermal anomaly is classified. Examples of classifications of thermal anomalies include: stationary targets, gas flares, agricultural anomalies, and wildfires. An action is taken based at least in part on the classified detected anomaly.

Abstract: Sparse phase unwrapping is disclosed. A first image and a second image are received. The first image and the second image are coregistered. The first image and the second image comprise respective phase data. An unwrapped interferogram is generated, including by solving an optimization problem using a nonconvex penalty function, where minimizing the penalty function produces sparse minimizers.

Abstract: Performing a geo-visual search is disclosed. A query feature vector associated with a query tile is obtained. A lookup is performed at least in part by using a key derived from the query feature vector. A list of candidate feature vectors is obtained based at least in part on the lookup. Based at least in part on a comparison of the query feature vector against at least some of the candidate feature vectors in the obtained list, a tile that is visually similar to the query tile is determined. The determined tile is provided as output.

Abstract: A set of images is received. A thermal anomaly is detected in the received set of images. The detected thermal anomaly is classified. Examples of classifications of thermal anomalies include: stationary targets, gas flares, agricultural anomalies, and wildfires. An action is taken based at least in part on the classified detected anomaly.

Abstract: A boundary map and a first image are received. The boundary map is used to determine that brightness values for a set of pixels included in the first image should be regularized. The first set of pixels include, at a first pixel position, a first pixel having a first set of brightness values. The first set of pixels further includes, at a second pixel position, a second set of brightness values. An output image is generated by storing, at both a first and second pixel position in the output image, a set of regularized values.

Abstract: Sparse phase unwrapping is disclosed. A first image and a second image are received. The first image and the second image are coregistered. The first image and the second image comprise respective phase data. An unwrapped interferogram is generated, including by solving an optimization problem using a nonconvex penalty function, where minimizing the penalty function produces sparse minimizers.

Abstract: Processing and analyzing observational data is disclosed. A set of data comprising at least a first tile and second tile is received. A first determination is made, for a first pixel position, which tile has a pixel with the highest value for that first pixel position. A second determination is made, for a second pixel position, which tile has a pixel with the highest value for that second pixel position. A composite image is generated using data obtained from the respective tiles in accordance with the two determinations.

Abstract: A first image is received. An initial label is assigned to at least some pixels in the first image, including by assigning a first label to a first pixel. A determination is made that the first pixel's label should be replaced with a different label. The first pixel's label is updated with the different label.

Abstract: A boundary map and a first image are received as input. The boundary map indicates a plurality of fields. The boundary map is used to regularize, across a set comprising a plurality of pixels bounded by a first field in the plurality of fields, a set of one or more pixel brightness values. A regularized image is generated as output, where for each pixel in the first image bounded by the first field, a set of values in the corresponding pixel in the regularized image is assigned the set of regularized pixel brightness values.

Abstract: A first image is received. An initial label is assigned to at least some pixels in the first image, including by assigning a first label to a first pixel. A determination is made, based at least in part on labels of spectral neighbors of the first pixel, that the first pixel's label should be replaced with a different label. The first pixel's label is updated with the different label. The first pixel's label is iteratively refined until convergence.

Abstract: Processing and analyzing observational data is disclosed. A set of data comprising at least a first tile and second tile is received. A first determination is made, for a first pixel position, which tile has a pixel with the highest value for that first pixel position. A second determination is made, for a second pixel position, which tile has a pixel with the highest value for that second pixel position. A composite image is generated using data obtained from the respective tiles in accordance with the two determinations.

Abstract: Performing a geo-visual search is disclosed. A query feature vector associated with a query tile is obtained. Based at least in part on a comparison of the query feature vector against at least some of a plurality of exemplar feature vectors, an exemplar feature vector is selected from the plurality of exemplar feature vectors. A list of candidate feature vectors associated with the selected exemplar feature vector is obtained. Based at least in part on a comparison of the query feature vector against at least some of the candidate feature vectors in the obtained list, a tile that is visually similar to the query tile is determined. The determined tile is provided as output.

Abstract: Various approaches to image enhancement are disclosed. In one approach, a boundary map and an image are received. The boundary map is used to determine that brightness values for a set of pixels included in the image should be regularized. An output image is generated by storing, at both a first and second pixel position in the output image, a set of regularized values. In another approach, an image to be enhanced and an edge map are received. Edge-masked derivative matrices are constructed and used to iteratively solve a series of linear equations, wherein solutions to the linear equations minimize an increasingly accurate quadratic approximation of a penalty function that measures a total amount of variation in a function, and a derivation of the function from the image. A vector result of the iterative solution is transformed into a raster image.

Abstract: A first image is received. An initial label is assigned to at least some pixels in the first image, including by assigning a first label to a first pixel. A determination is made, based at least in part on labels of spectral neighbors of the first pixel, that the first pixel's label should be replaced with a different label. The first pixel's label is updated with the different label.

Abstract: A set of data comprising at least a first tile and a second tile is received. A determination of a measure is made, for a first pixel location included in both the first tile and the second tile, of whether the first pixel location corresponds to a boundary. A boundary map is generated as output. The boundary map can be used, in conjunction with an image, to regularize pixel brightness values of the image.

Abstract: Various approaches to image enhancement are disclosed. In one approach, an image to be enhanced and a boundary map are received. A field value array is made using the boundary map and the received image. For each field represented in the field value array, pixel values obtained from the received image are aggregated. An enhanced image is generated as output by performing a rereasterization using the aggregated values. In another approach, an image to be enhanced and an edge map are received. Edge-masked derivative matrices are constructed. The edge-masked derivative matrices are used to iteratively solve a series of linear equations, wherein solutions to the linear equations minimize an increasingly accurate quadratic approximation of a penalty function that measures a total amount of variation in a function, and a derivation of the function from the image. A vector result of the iterative solution is transformed into a raster image.

Abstract: A first image is received. An initial label is assigned to at least some pixels in the first image, including by assigning a first label to a first pixel. A determination is made, based at least in part on labels of spectral neighbors of the first pixel, that the first pixel's label should be replaced with a different label. The first pixel's label is updated with the different label.