| CPC A01B 79/005 (2013.01) [G06N 20/00 (2019.01); G06Q 50/02 (2013.01); G06V 10/758 (2022.01); G06V 20/188 (2022.01)] | 19 Claims |
|
1. A method comprising:
accessing, by a hardware processor, tile data describing a plurality of tile images for a geographic region over a period of time, the tile data comprising pixels and being generated based on satellite remote sensing data;
generating, by the hardware processor and based on the tile data, farm field pixel data for a farm field of interest, the farm field pixel data describing a plurality of pixels for the farm field of interest over the period of time;
generating, by the hardware processor and based on the farm field pixel data, vegetation index data for the farm field of interest by, for each individual pixel in the plurality of pixels:
generating an initial time series of vegetation index values for the individual pixel over the period of time; and
generating a reconstructed time series of vegetation index values for the individual pixel over the period of time by using a first machine learning model to generate a predicted vegetation index value for each vegetation index value in the initial time series of vegetation index values, the first machine learning model comprising a neural network model trained to reconstruct vegetation index values of all pixels of a field;
generating, by the hardware processor, a crop growth curve for the farm field of interest based on the vegetation index data;
generating, by the hardware processor, a crop growth band based on historical data that describes vegetation index values for one or more other farm fields in the geographic region;
causing, by the hardware processor, presentation of the crop growth band and the crop growth curve on a client graphical user interface on a client device, the crop growth band and the crop growth curve being presented relative to each other for comparison;
generating, using a second machine learning model, a simulated yield value for the field of interest based on the crop growth band; and
generating, by the hardware processor, one or more alerts based on the simulated yield value for the field of interest, the one or more alerts being reviewable by a user, at least one alert of the one or more alerts providing actionable information to the user for adjusting management of the field of interest.
|