Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for image segmentation and chemical analysis using machine learning. In some implementations, a system obtains a hyperspectral image that includes a representation of an object. The system segments the hyperspectral image to identify regions of a particular type on the object. The system generates a set of feature values derived from image data for different wavelength bands that is located in the hyperspectral image in the identified regions of the particular type. The system generates a prediction of a level of one or more chemicals in the object based on an output produced by a machine learning model in response to the set of feature values being provided as input to the machine learning model. The system provides data indicating the prediction of the level of the one or more chemicals in the object.

The present disclosure provides a system for monitoring unstructured environments. A predetermined path can be determined according to an assignment of geolocations to one or more agronomically anomalous target areas, where the one or more agronomically anomalous target areas are determined according to an analysis of a plurality of first images that automatically identifies a target area that deviates from a determination of an average of the plurality of first images that represents an anomalous place within a predetermined area, where the plurality of first images of the predetermined area are captured by a camera during a flight over the predetermined area. A camera of an unmanned vehicle can capture at least one second image of the one or more agronomically anomalous target areas as the unmanned vehicle travels along the predetermined path.

A method of generating a roadway map includes receiving an image of a roadway. The method further includes performing a spectral analysis of the received image to determine reflectivity data for a plurality of wavelengths of light. The method further includes identifying a feature of the roadway in response to the determined reflectivity data exhibiting a reflection peak. The method further includes classifying the identified feature based on a size or a pitch of the exhibited reflection peak. The method further includes generating the roadway map based on the classification of the identified feature.

There is provided a digital map product comprising data informative of a location of a zone in a surface area including a subsurface utility infrastructure (SUI), the digital map product being derivative of a method comprising: receiving a digital image of the surface area; identifying a plurality of surface features; for each of the plurality of surface features: calculating an indication of utility location (IUL) from the respective surface feature and location, wherein the IUL is one of: a location of a point of the SUI, a location of a zone of the SUI, a location of a zone from which the SUI is absent, thereby giving rise of a plurality of IULs; and defining a location of a zone including the SUI in accordance with, at least, the plurality of IULs wherein at least one surface feature of the plurality of surface features is a public works surface marking.

A large-scale forest height remote sensing retrieval method includes: acquiring Ice, Cloud and land Elevation Satellite (ICESAT-2) tree height data, Landsat data, Shuttle Radar Topography Mission (SRTM) data, Worldclim data, forest type data and ecological zoning data within a target zone, and preprocessing the data; carrying out georeferencing on the processed data to generate a first data set; calculating spectral features, terrain features and climatic factor features of an image, and combining the calculated features with the ecological zoning data and the forest type data to obtain a second data set; extracting eigenvalues of a same geographical location from the second data set, and combining the extracted eigenvalues with the tree height data to generate training data; constructing a random forest model covering a large zone as an ecological zoning tree height retrieval model, and dividing the obtained training data into a training sample and a verification sample.

A few-shot urban remote sensing image information extraction method based on meta learning and attention includes building a few-shot urban remote sensing information pre-trained model. During a pre-training stage, pre-training network learning is performed for a few-shot set to fully learn feature information of existing samples and obtain initial feature parameters and a deep convolutional network backbone of the few-shot set; the few-shot urban remote sensing information pre-trained model is a network structure including a convolutional layer, a pooling layer and a fully-connected layer, and includes five sections of convolutional network where each section includes two or three convolutional layers, and an end of each section is connected to one maximum pooling layer to reduce a size of a picture; the number of convolutional kernels inside each section is same, and when closer to the fully-connected layer, the number of convolutional kernels is larger.

The present disclosure discloses a remote sensing-based extraction method for a type of a river channel, including: obtaining multi-source remote sensing data of a target area including a river channel; preprocessing the multi-source remote sensing data, and obtaining corresponding reflectance data; according to the reflectance data, analyzing a water index and a vegetation index of the target area; and according to the water index and the vegetation index, constructing first preset conditions for determining whether there is water in the river channel and second preset conditions for determining whether the river channel is a non-dry river channel to determine the type of the river channel. Types of river channels can be divided into four types: non-dry river channels, seasonal dry river channels, temporary water channels, and dry river channels.

A method and a system for object detection over water are provided. The method includes acquiring, from a storage system, image data associated with a target area. The image data includes radiance of a plurality of pixels in a first spectral band and in a second spectral band. The method also includes determining a metric corresponding to a pixel of the plurality of pixels as a function of the radiance in the first spectral band and the radiance in the second spectral band and detecting an object in the target area in response to a determination that the metric satisfies a criteria.

The present disclosure provides a method, apparatus, medium and device for extracting a river drying-up region and frequency, and belongs to the technical field of remote sensing. The method for extracting a river drying-up region and frequency can be applied to efficiently acquiring river drying-up information for a long time within the large spatial region. With the inverse normalized difference water index (iNDWI) and the maximum value composite (MVC) method for the remote sensing images, the present disclosure omits the troublesome step of separately extracting a river range for each image in the conventional method. In addition, the present disclosure quickly obtains the drying-up frequency by counting the total number of available images and the number of non-water images at the same pixel position, and yields a greater efficiency for extracting the river drying-up region and frequency.

A method for optimizing image data for generating orthorectified image(s) related to an area of interest in an environment. The method includes receiving a first image dataset of the area of interest captured therein, identifying each of multiple objects in the area of interest, receiving attribute information related to each of the multiple identified objects, determining if one or more of the multiple identified objects satisfy at least one of a risk criteria based on the attribute information therefor, identifying a maximum relevant second area including at least the area of interest and each of the one or more of the multiple identified objects satisfying the at least one of risk criteria, and processing the first image dataset to either discard or down-sample areas other than the maximum relevant second area captured therein.