Crop scouting information systems and resource management
Described herein are techniques for generating contextually rich plant images. A number of data captures of raw plant data are generated via a sensing unit configured to navigate a growing facility. Metadata is generated and assigned to the raw plant data including at least one of: plant location, timestamp, plant identification, plant strain, facility identification, facility location, facility type, health risk factors, plant conditions, and human observations. Images generated by the sensing unit are analyzed and pixel annotations are generated in the images based on their relation to one or more plant well-being features. Data tags are generated and assigned the data captures based on an analysis of the data captures. The data tags are text phrases linking a particular data capture to a specific threat to plant well-being.
1. A method of generating contextually rich plant images comprising:
generating a plurality of data captures of raw plant data via a sensing unit configured to navigate rows of plants within a growing facility;
generating and assigning metadata to the raw plant data including at least one of: plant location, timestamp, plant identification, plant strain, facility identification, facility location, facility type, health risk factors, plant conditions, and human observations;
analyzing images generated by the sensing unit and generating pixel annotations in the images based on their relation to one or more plant well-being features;
assigning data tags to one or more of the plurality of data captures based on an analysis of the plurality of data captures, wherein the data tags are text phrases linking a particular data capture to a specific threat to plant well-being;
performing image processing on the images generated by the sensing unit, the image processing including at least one of:
transforming an image by cropping the image into a smaller image, reducing noise, modifying sharpness or brightness, or performing color balancing;
stitching two or more images together to generate a panorama image to minimize double counting plant regions;
overlaying different images of a single scene collected with different cameras; or
cropping a panorama image to fit a particular aspect ratio;
generating a database query based on user input received from a user interface;
comparing user input received from the user interface against the data tags assigned to one or more of the plurality of data captures; and
generating a curated data set including a subset of the plurality of data captures corresponding to a combination of the tags, based on the comparison with the user input.
2. The method of claim 1 , wherein the plurality of data captures includes a collection of images across visible and non-visible light spectra.
3. The method of claim 1 , wherein the plurality of data captures includes a collection of thermal images.
4. The method of claim 1 , wherein the plurality of data captures includes a collection of environmental readings comprising at least one of: temperature, humidity, luminosity, radiation, magnetic field, particulate matter, and chemical compounds.
5. The method of claim 1 , wherein the plurality of data captures includes a collection of contextual readings comprising at least one of: acceleration, gyroscope position, orientation, previous system state, next planned system state, power level, CPU usage, CPU memory, and communication signal.
6. The method of claim 4 , further comprising:
generating a database containing:
the raw plant data from the plurality of data captures;
the metadata assigned to the raw plant data;
the pixel annotations in the images; and
the data tags assigned to one or more of the plurality of data captures.
7. The method of claim 1 , further comprising:
generating a plant profile for one or more plants within the growing facility, wherein each plant profile includes:
a plant identifier identifying a particular plant;
sensor information generated by the sensing unit related to the particular plant;
metadata related to raw plant data of the particular plant; and
data tags related to raw plant data of the particular plant.
8. A method of training a machine learning model comprising:
querying a database containing raw plant data from a plurality of data captures, metadata assigned to the raw plant data, pixel annotations in images analyzed from the plurality of data captures, and data tags assigned to one or more of the plurality of data captures wherein the data tags are text phrases linking a particular data capture to a specific threat to plant well-being, wherein the database query includes at least one tag or data parameter;
generating a curated training data set including a subset of the plurality of data captures corresponding to the at least one tag or data parameter;
selecting one or more features from the pixel annotations in images analyzed from the plurality of data captures;
building a trained machine learning model based on an analysis of the curated training data set and the one or more features, wherein the trained machine learning model is configured to:
receive raw plant data from the plurality of data captures;
associate a subset of the raw plant data with the one or more features; and
identify an existence of one or more plant abnormalities within the subset of the raw plant data based on the association of the subset with the one or more features;
analyzing raw plant data from previous data captures associated with a particular plant to associate earlier data features with the one or more plant abnormalities when the trained machine learning model identifies the existence of the one or more plant abnormalities within the particular plant, wherein analyzing raw plant data from previous data captures comprises overlaying images of the particular plant from different time intervals to detect a pattern in raw plant data prior to detecting a visible plant abnormality.
9. The method of claim 8 , wherein the at least one tag or data parameter includes a text phrase related to a specific threat to plant well-being, such as disease, insects, pest activity, dehydration, nutrient deficiencies, future diseases, harvest yield, or harvest time.
10. The method of claim 8 , wherein the images analyzed from the plurality of data captures include thermal images depicting temperature variations within one or more plants, and the trained machine learning model is configured to identify the existence of one or more plant abnormalities within the subset of the raw plant data based on the temperature variations.
11. The method of claim 8 , wherein the trained machine learning model is an artificial neural network comprising a plurality of input nodes, one or more hidden layers, and a plurality of output nodes, wherein each input node includes a memory location for storing input values including the raw plant data from the plurality of data captures.
12. The method of claim 8 , further comprising generating a plurality of risk scores corresponding to the one or more plant abnormalities.