Estimation of weed population dynamics in fields or other growing areas

Abstract

Satellite images (900) of a growing area captured repeatedly within a specified time window are obtained (602, 702). At least some of the satellite images are selected (706) for processing, one or more overall characteristics of the growing area at one or more times are identified (802) using at least some of the selected satellite images, and a subset of the selected satellite images within the specified time window is selected (806) based on the overall characteristic(s). The growing area is divided (1002) into multiple regions (904), and one or more characteristics of each of the regions of the growing area are identified (1004) based on the subset of the selected satellite images. At least one classifier (306) is used to classify (1006, 1010) each of the regions of the growing area as either containing weeds or not containing weeds based on the one or more characteristics of the regions of the growing area and the one or more overall characteristics of the growing area.

Description

GECO01-00006 1 ESTIMATION OF WEED POPULATION DYNAMICS IN FIELDS OR OTHER GROWING AREAS TECHNICAL FIELD

[0001] This disclosure is generally directed to prediction systems. More specifically, this disclosure is directed to estimation of weed population dynamics in fields or other growing areas. BACKGROUND

[0002] Identifying actual or expected locations of weeds in a field or other growing area can be useful or beneficial for various reasons. For example, chemical herbicides are a primary tool for the control of weeds in modern agricultural production. In many farm fields or other growing areas, weeds often grow in patches, and the patches may be located anywhere within the growing areas. Common locations of weed patches may include around the edges of a field or other growing area or in areas where soil is locally more favorable to weeds than crops, such as when the weed kochia tends to be advantaged relative to most crops in saline and dry soil areas or when weed seeds have accumulated due to environmental, human, or animal factors. Chemical herbicides are a common and very effective tool for controlling weeds. Ideally, chemical herbicides would be applied only at the actual or expected locations of weeds in a field or other growing area in order to reduce or minimize the usage of the chemical herbicides. SUMMARY

[0003] This disclosure relates to estimation of weed population dynamics in fields or other growing areas.

[0004] In a first embodiment, a method includes obtaining satellite images of a growing area, where the satellite images are captured repeatedly within a specified time window. The method also includes selecting at least some of the satellite images for processing and identifying one or more overall characteristics of the growing area at one or more times using at least some of the selected satellite images. The method further includes selecting a subset of the selected satellite images within the specified time window based on the one or more overall characteristics and dividing the growing area into multiple regions. The method also includes identifying one or more characteristics of each of the regions of the growing area based on the subset of the selected satellite images. In addition, the method includes using at least one classifier to classify each of the regions of the growing area as either containing weeds or not containing weeds based on the one or more characteristics of the regions of the growing area and the one or more overall characteristics of theGECO01-00006 2 growing area.

[0005] In a second embodiment, an apparatus includes at least one processing device configured to obtain satellite images of a growing area, where the satellite images are captured repeatedly within a specified time window. The at least one processing device is also configured to select at least some of the satellite images for processing and identify one or more overall characteristics of the growing area at one or more times using at least some of the selected satellite images. The at least one processing device is further configured to select a subset of the selected satellite images within the specified time window based on the one or more overall characteristics. The at least one processing device is also configured to divide the growing area into multiple regions and identify one or more characteristics of each of the regions of the growing area based on the subset of the selected satellite images. In addition, the at least one processing device is configured to use at least one classifier to classify each of the regions of the growing area as either containing weeds or not containing weeds based on the one or more characteristics of the regions of the growing area and the one or more overall characteristics of the growing area.

[0006] In a third embodiment, a non-transitory computer readable medium stores computer readable program code that when executed causes at least one processor to obtain satellite images of a growing area, where the satellite images are captured repeatedly within a specified time window. The non-transitory computer readable medium also stores computer readable program code that when executed causes the at least one processor to select at least some of the satellite images for processing and identify one or more overall characteristics of the growing area at one or more times using at least some of the selected satellite images. The non-transitory computer readable medium further stores computer readable program code that when executed causes the at least one processor to select a subset of the selected satellite images within the specified time window based on the one or more overall characteristics. The non-transitory computer readable medium also stores computer readable program code that when executed causes the at least one processor to divide the growing area into multiple regions and identify one or more characteristics of each of the regions of the growing area based on the subset of the selected satellite images. In addition, the non-transitory computer readable medium stores computer readable program code that when executed causes the at least one processor to use at least one classifier to classify each of the regions of the growing area as either containing weeds or not containing weeds based on the one or more characteristics of the regions of the growing area and the one or more overall characteristics of the growing area.

[0007] Other technical features may be readily apparent to one skilled in the art from theGECO01-00006 3 following figures, descriptions, and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] For a more complete understanding of this disclosure, reference is now made to the following description, taken in conjunction with the accompanying drawings, in which:

[0009] FIGURE 1 illustrates an example system supporting the estimation of weed population dynamics in a field or other growing area according to this disclosure;

[0010] FIGURE 2 illustrates an example computing device supporting the estimation of weed population dynamics in a field or other growing area according to this disclosure;

[0011] FIGURE 3 illustrates an example architecture supporting the estimation of weed population dynamics in a field or other growing area according to this disclosure;

[0012] FIGURE 4 illustrates an example aggregation of weed locations to support the estimation of weed population dynamics in a field or other growing area according to this disclosure;

[0013] FIGURE 5 illustrates example historical data used for estimation of weed population dynamics in a field or other growing area according to this disclosure;

[0014] FIGURE 6 illustrates an example method for estimating weed population dynamics in a field or other growing area according to this disclosure;

[0015] FIGURE 7 illustrates an example method for obtaining information to be used for estimating weed population dynamics in a field or other growing area according to this disclosure;

[0016] FIGURE 8 illustrates an example method for processing information for estimating weed population dynamics in a field or other growing area according to this disclosure;

[0017] FIGURE 9 illustrates an example grid-based approach for processing information for estimating weed population dynamics in a field or other growing area according to this disclosure; and

[0018] FIGURE 10 illustrates an example method for classification to support estimation of weed population dynamics in a field or other growing area according to this disclosure. DETAILED DESCRIPTION

[0019] FIGURES 1 through 10, described below, and the various embodiments used to describe the principles of the present invention in this patent document are by way of illustration only and should not be construed in any way to limit the scope of the invention. Those skilled in the art will understand that the principles of the present invention may be implemented in any typeGECO01-00006 4 of suitably arranged device or system.

[0020] As noted above, identifying actual or expected locations of weeds in a field or other growing area can be useful or beneficial for various reasons. For example, chemical herbicides are a primary tool for the control of weeds in modern agricultural production. In many farm fields or other growing areas, weeds often grow in patches, and the patches may be located anywhere within the growing areas. Common locations of weed patches may include around the edges of a field or other growing area or in areas where soil is locally more favorable to weeds than crops, such as when the weed kochia tends to be advantaged relative to most crops in saline and dry soil areas or when weed seeds have accumulated due to environmental, human, or animal factors. Chemical herbicides are a common and very effective tool for controlling weeds. Ideally, chemical herbicides would be applied only at the actual or expected locations of weeds in a field or other growing area in order to reduce or minimize the usage of the chemical herbicides.

[0021] In general, identifying actual weed emergence and predicting future weed emergence so that only certain parts of a growing area are treated with herbicide is difficult and depends on many factors each growing season. These factors may include complex interactions of variables, such as weather conditions, crop competitiveness, and soil composition. With respect to weather conditions, factors such as temperature and moisture may need to be just right for some weed seeds to germinate, while other weed seeds may germinate over a wide range of conditions. With respect to crop competitiveness, some crops may be more competitive with weeds than others, meaning competition is a function of specific weed-crop combinations. With respect to soil composition, factors such as salinity can affect how well certain crops or certain weeds grow. As a result, weeds tend not to emerge all at once and can instead emerge all throughout a growing season depending on factors like weed species, weather, and crop species. As a particular example of this, the weed kochia has a wide window of emergence due to its unique ability to germinate across a range of temperatures (including temperatures in early January in some locations) and may continue to emerge throughout the summer and into fall. High seedling emergence in early spring and continued emergence into mid- or late-summer or fall means kochia often needs to be managed for a prolonged period of time. Kochia seeds do not generally exhibit innate dormancy, so mature seeds often germinate rapidly under favorable conditions, which prolongs their emergence period.

[0022] This disclosure describes various techniques supporting the estimation of weed population dynamics in fields or other growing areas. For example, this disclosure describes techniques in which lower-resolution but higher-frequency data (such as images of a field or other growing area from one or more satellites) over a desired time window is processed. In some cases,GECO01-00006 5 the time window may represent a prolonged period of time, such as five years or even longer. Also, in some cases, the lower-resolution but higher-frequency data may be combined with higher- resolution but lower-frequency data (such as images of the field or other growing area from one or more higher-resolution satellites, drones, automated sprayers, or other devices). Filtering may be performed to remove images or portions thereof from further processing, such as images or parts of images in which the ground or plants are obscured by clouds or snow. The remaining images can be processed, such as by identifying one or more characteristics of individual regions of the growing area (such as grid sections of the growing area) and for the overall growing area. One or more machine learning-based classifiers or other classification algorithms can be applied to the image processing results in order to classify whether any portions of the growing area appear to contain weeds (and optionally what kind of weeds might be present, if any, or what absolute or relative density those weeds might be at, if any). In some cases, the classification results can be used to generate one or more weed maps, where each weed map identifies locations in the growing area where weeds are present or might be present in the future. Also, in some cases, additional processing may be performed on the classifications or the resulting weed maps, such as when spatial clustering is performed to create smoother weed maps (like weed maps that do not follow grid lines).

[0023] Weed maps or other results of this process may be used in any suitable manner. For example, weed maps or other results may be used to determine which portions of the growing area should be treated with a chemical herbicide or other treatment, such as when the weed maps or other results are provided to or used with one or more tractor-based or other land-based sprayers or one or more drone-based or other aerial-based sprayers. The weed maps or other results may be combined with estimates of weed locations produced by one or more tractor-based or other land- based sprayers or surveyors or by one or more drone-based or other aerial-based sprayers or surveyors, such as to augment estimates of weed locations and reduce false negatives or false positives. Weed maps or other results may be used to estimate the locations of weed seedbanks within the growing area. A “weed seedbank” or “seedbank” refers to at least one collection of weed seeds that could potentially germinate and produce weeds within a growing area. A weed seedbank typically (but not necessarily) is associated with seeds that are underground and waiting for the right conditions to germinate. As a result, weed seedbanks are typically not detectable to the naked eye and are often only discovered by human personnel after weeds have germinated. Weed maps or other results may be used to estimate whether weeds in a particular growing area are showing signs of herbicide resistance. For instance, weed maps before and after application of an herbicideGECO01-00006 6 can be compared to see if the herbicide was effective at killing all of the weeds in a treated area or if at least some of the weeds remain (indicating potential resistance to the herbicide). If potential herbicide resistance is detected, this information may be used to take additional action, such as initiating application of a different herbicide to kill the remaining weeds that may be resistant to the original herbicide. Note, however, that these examples of use cases are for illustration only. In general, estimated weed population dynamics in fields or other growing areas may be used in any other suitable manner, and this disclosure is not limited to any specific use cases for the estimated weed population dynamics.

[0024] One example use of the techniques described below involves treating portions of growing areas that have or are predicted to have weeds. In many instances, the use of an herbicide is often described in this document as involving spraying of the herbicide. However, herbicides may be deployed in any suitable manner. For example, some herbicides have a solid form, such as when the herbicides are applied in granular form. As a result, various types of equipment may be used to apply one or more herbicides, such as one or more sprayers, granular applicators, or seed drills. While sprayers are often described below, it will be understood that any suitable mechanisms can be used to deliver herbicides in any suitable form. Also, other types of treatments may be used along with or instead of herbicides. Examples of various types of treatments that may be applied to crops could include a multi-rate application of an herbicide, a multi-herbicide application of multiple herbicides, an increase in seeding density for crops, a change in crop, a blanket application of herbicide, and a targeted application of one or more nutrients and / or fertilizer. The terms “treatment” and “treatments” are used in this document to encompass one or more actions (whether preventative or remedial) that can reduce the number or presence of weeds in at least one growing area.

[0025] FIGURE 1 illustrates an example system 100 supporting the estimation of weed population dynamics in a field or other growing area according to this disclosure. As shown in FIGURE 1, the system 100 includes user devices 102a-102d, one or more networks 104, one or more application servers 106, and one or more database servers 108 associated with one or more databases 110. Each user device 102a-102d communicates over the network 104, such as via a wired or wireless connection. Each user device 102a-102d represents any suitable device or system used by at least one user to provide or receive information, such as a desktop computer, a laptop computer, a smartphone, and a tablet computer. However, any other or additional types of user devices may be used in the system 100. In some cases, one or more users may use one or more user devices 102a-102d to identify weeds in at least one growing area. In other cases, one or more usersGECO01-00006 7 may use one or more user devices 102a-102d to view a graphical user interface or other interface that presents analysis results (such as an estimation of weed population dynamics) and trigger any suitable actions (such as scheduling or approving herbicide application or other treatments in areas containing or at risk for weeds).

[0026] The network 104 facilitates communication between various components of the system 100. For example, the network 104 may communicate Internet Protocol (IP) packets, frame relay frames, Asynchronous Transfer Mode (ATM) cells, or other suitable information between network addresses. The network 104 may include one or more local area networks (LANs), metropolitan area networks (MANs), wide area networks (WANs), all or a portion of a global network such as the Internet, or any other communication system or systems at one or more locations. In some cases, the network 104 may represent a combination of networks. For instance, the one or more user devices 102a-102d may communicate over a local area network, and the one or more application servers 106 and the one or more database servers 108 may be remote (possibly located within a cloud-based environment) and may communicate with the local area network over a wide area network or global network.

[0027] The application server 106 is coupled to the network 104 and is coupled to or otherwise communicates with the database server 108. The application server 106 supports the analysis of data (which may be obtained from one or more data sources 114 and stored in the database 110) in order to estimate weed population dynamics in one or more growing areas. Example operations that may be performed by the application server 106 are described below. In some embodiments, the application server 106 may execute one or more applications 112 that use data from the database 110 to estimate the weed population dynamics. Note that the database server 108 may also be used within the application server 106 to store information, in which case the application server 106 itself may store the information used to predict the weed population dynamics.

[0028] The database server 108 operates to store and facilitate retrieval of various information used, generated, collected, or provided by the application server 106, the user devices 102a-102d, the data sources 114, and / or other components in the database 110. For example, the database server 108 may store various information related to vegetation or other information related to weeds or other plants detected in one or more growing areas. In some cases, the database server 108 may store information collected over a prolonged period of time for a growing area, such as up to five years of data for the growing area or even more.

[0029] The one or more data sources 114 may represent any suitable source(s) of data analyzed by the application server 106 to estimate weed population dynamics. For example, the one or moreGECO01-00006 8 data sources 114 may include one or more sources of satellite images or other satellite-based data or other remotely-sensed data associated with at least one field or other growing area. In some cases, the satellite-based data may include multi-spectral data. As a particular example, the satellite- based data may include normalized difference vegetation index (NDVI) data. The one or more data sources 114 may optionally include one or more sources of image data or other data captured using at least one smart spraying system or other smart herbicide application system, such as data captured using cameras or other imaging sensors on one or more tractors, all-terrain vehicles (ATVs), airborne drones, or other vehicles that are equipped with systems for selectively spraying weeds or otherwise applying herbicide. The one or more data sources 114 may optionally include one or more sources of image data or other data captured using at least one surveying device, such as data captured using cameras or other imaging sensors on one or more tractors, ATVs, drones, or other vehicles designed to provide surveying (but not herbicide application) capabilities. Note, however, that any other suitable source(s) of data may be used here. For instance, data sources used for estimation of weed population dynamics may include one or more agronomically-relevant data sources, such as one or more sources of human-collected scouting data, meteorological data, soil type data, weed species data, data associated with prior herbicide applications, and data defining management practices. The human-collected scouting data may include locations of weeds as identified by human personnel scouting a growing area.

[0030] One or more automated platforms 116 may optionally be used in the system 100. In some cases, the one or more automated platforms 116 may include one or more platforms that can identify weeds in one or more growing areas. For example, the one or more automated platforms 116 may include tractors, ATVs, drones, or other devices configured to identify weeds during a survey or other operations. The one or more automated platforms 116 may also or alternatively include one or more camera-enabled or other sensor-enabled smart spraying systems or other herbicide application systems, such as tractors, ATVs, drones, or other devices configured to apply treatments to weeds while trying to avoid treating other plants like crops. As a result, the same device may represent both a data source 114 and an automated platform 116. However, an automated platform 116 may also represent a platform that does not function as a data source 114, such as when an automated platform 116 represents a tractor-based, ATV-based, drone-based, or other spraying system or other herbicide application system that does not differentiate between weeds and other plants. One or more of the automated platforms 116 may optionally be controlled based on estimations of weed population dynamics, such as when at least one smart or other tractor- based, ATV-based, drone-based, or other spraying system or other herbicide application systemGECO01-00006 9 can be controlled to apply herbicide at locations where weeds are identified or expected within a growing area.

[0031] Although FIGURE 1 illustrates one example of a system 100 supporting the estimation of weed population dynamics in a field or other growing area, various changes may be made to FIGURE 1. For example, various components shown in FIGURE 1 may be combined, further subdivided, replicated, omitted, or rearranged and additional components may be added according to particular needs. Also, the system 100 may include any number of user devices 102a-102d, networks 104, application servers 106, database servers 108, databases 110, data sources 114, and automated platforms 116 (possibly including zero of one or more of these components). Further, these components may be located in any suitable locations and might be distributed over a large area. In addition, while FIGURE 1 illustrates one example operational environment in which the estimation of weed population dynamics in a field or other growing area may be used, this functionality may be used in any other suitable system.

[0032] FIGURE 2 illustrates an example computing device 200 supporting the estimation of weed population dynamics in a field or other growing area according to this disclosure. One or more instances of the device 200 may, for example, be used to at least partially implement the functionality of the application server 106 shown in FIGURE 1. However, the functionality of the application server 106 may be implemented in any other suitable manner. In some embodiments, the device 200 shown in FIGURE 2 may form at least part of a user device 102a-102d, application server 106, database server 108, data source 114, or automated platform 116 in FIGURE 1. However, each of these components may be implemented in any other suitable manner.

[0033] As shown in FIGURE 2, the device 200 denotes a computing device or system that includes at least one processing device 202, at least one storage device 204, at least one communications unit 206, and at least one input / output (I / O) unit 208. The processing device 202 may execute instructions that can be loaded into a memory 210. In some embodiments, the processing device 202 may execute instructions to estimate weed population dynamics in a field or other growing area. The processing device 202 may also optionally execute instructions to generate recommendations or trigger treatments in response to the estimations. Examples of the types of functions that may be performed using the processing device 202 are provided below. The processing device 202 includes any suitable number(s) and type(s) of processors or other processing devices in any suitable arrangement. Example types of processing devices 202 include one or more microprocessors, microcontrollers, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or discrete circuitry.GECO01-00006 10

[0034] The memory 210 and a persistent storage 212 are examples of storage devices 204, which represent any structure(s) capable of storing and facilitating retrieval of information (such as data, program code, and / or other suitable information on a temporary or permanent basis). The memory 210 may represent a random access memory or any other suitable volatile or non-volatile storage device(s). The persistent storage 212 may contain one or more components or devices supporting longer-term storage of data, such as a read only memory, hard drive, Flash memory, or optical disc.

[0035] The communications unit 206 supports communications with other systems or devices. For example, the communications unit 206 can include a network interface card or a wireless transceiver facilitating communications over a wired or wireless network, such as the network 104. The communications unit 206 may support communications through any suitable physical or wireless communication link(s).

[0036] The I / O unit 208 allows for input and output of data. For example, the I / O unit 208 may provide a connection for user input through a keyboard, mouse, keypad, touchscreen, or other suitable input device. The I / O unit 208 may also send output to a display, printer, or other suitable output device. Note, however, that the I / O unit 208 may be omitted if the device 200 does not require local I / O, such as when the device 200 represents a server or other device that can be accessed remotely.

[0037] Although FIGURE 2 illustrates one example of a computing device 200 supporting the estimation of weed population dynamics in a field or other growing area, various changes may be made to FIGURE 2. For example, various components shown in FIGURE 2 may be combined, further subdivided, replicated, omitted, or rearranged and additional components may be added according to particular needs. Also, computing and communication devices and systems come in a wide variety of configurations, and FIGURE 2 does not limit this disclosure to any particular computing or communication device or system.

[0038] FIGURE 3 illustrates an example architecture 300 supporting the estimation of weed population dynamics in a field or other growing area according to this disclosure. For ease of explanation, the architecture 300 shown in FIGURE 3 is described as being implemented using the application server 106 in the system 100 shown in FIGURE 1, where the application server 106 may be implemented using one or more instances of the device 200 shown in FIGURE 2. However, the architecture 300 may be implemented using any other suitable device(s) and in any other suitable system(s).

[0039] As shown in FIGURE 3, the architecture 300 includes or has access to one or moreGECO01-00006 11 data sources 302, which can provide information to be processed by the architecture 300. The one or more data sources 302 may include any suitable source(s) of relevant weed-related or plant- related data, such as the database 110 and / or the one or more data sources 114. The one or more data sources 302 may provide any suitable information to the architecture 300 for processing, such as various information related to vegetation or other information related to weeds or other plants in one or more growing areas. Specific examples can include satellite images or other satellite-based data or other remotely-sensed data (such as multi-spectral data or multi-spectral metrics like NDVI data), image data or other data captured using at least one smart herbicide application system, image data or other data captured using at least one surveying device, human-collected scouting data, meteorological data, soil type data, weed species data, data associated with prior herbicide applications, data defining management practices, or any combination thereof.

[0040] In some cases, the one or more data sources 302 can provide satellite images of one or more fields or other growing areas. Satellites may provide imagery of a field or other growing area more frequently (such as several times per week), and there are typically historical satellite images that are available for use. Satellite imagery also tends to be much less expensive compared to other types of imaging, and satellite imagery often provides 100% coverage of a given area. However, satellite imagery typically has a lower resolution (such as around three to ten meters) compared to other types of imaging. In contrast, land-based and aerial-based imaging systems can have much higher resolution (such as less than one millimeter to one meter) compared to satellite imagery. However, land-based and aerial-based imaging systems may provide images far less frequently (such as one to four times per year), are far more expensive, generally do not provide historical data, and are used with a very small number of fields or other growing areas. Thus, some embodiments of this disclosure may use satellite images only, and other embodiments of this disclosure may use satellite images combined with other data (such as from one or more land-based and aerial-based imaging systems) if that other data is available.

[0041] One or more data processing functions 304 receive the data from the data source(s) 302 and process the data in order to generate weed-related information associated with fields or other growing areas. For example, one or more data processing functions 304 may involve georeferencing data in order to associate specific data with one or more specific fields or other growing areas and identifying boundaries of the one or more fields or other growing areas. This allows the architecture 300 to identify which of the data being processed relates to which field or other growing area. As a specific example, the one or more data processing functions 304 may divide a field or other growing area into regions using a grid pattern and identify image data orGECO01-00006 12 other data that relates to different regions of the field or other growing area. The one or more data processing functions 304 can also process the image data or other data associated with fields or other growing areas to identify characteristics of the regions of each field or other growing area and of the overall field or other growing area based on the image data, such as by identifying an average NDVI value, an average normalized difference red edge (NDRE) index value, average red / green / blue (RGB) values, an average near infrared (NIR) value, or other value(s) or any combination thereof for each region of the field or other growing area and an average NDVI value or other value(s) or any combination thereof for the entire field or other growing area. Depending on the implementation, the images and other data processed here could span any suitable period of time, including a prolonged period of time (such as five years or more).

[0042] At least one machine learning-based or other weed classifier 306 can process the data from the one or more data processing functions 304 in order to determine actual or predicted locations of weeds in the fields or other growing areas. For example, the weed classifier 306 may select a subset of images from among the collection of images processed by the one or more data processing functions 304, where the subset of images may represent those images that are likely to show weed locations. In some cases, the weed classifier 306 may select the subset of images as those images captured within the relevant period of time having one or more average values within one or more ranges of desired values. The one or more average values could be associated with one or more characteristics of an overall field or other growing area, such as the average NDVI value for an entire field or other growing area. As a particular example, the weed classifier 306 may select the subset of images as those images captured within the relevant period of time having an overall NDVI value for the entire field or other growing area greater than 0.1 but less than 0.25. This may represent an overall condition of the field or other growing area that allows germinated weeds to be viewed without significant crop interference. Note, however, that any other suitable condition or conditions may be used here to select images.

[0043] For the selected subset of images, the weed classifier 306 can operate to classify locations within the field or other growing area as being associated with weeds or not being associated with weeds. In some cases, the weed classifier 306 can do this on a per-pixel basis, meaning the weed classifier 306 can determine whether or not each pixel of each image in the subset is or is not associated with weeds. This classification can be based on one or more characteristics of the overall field or other growing area and / or one or more characteristics of each region of the field or other growing area. As described below, for instance, the weed classifier 306 could normalize or standardize the NDVI values, NDRE values, RGB values, NIR values, or otherGECO01-00006 13 characteristic(s) of the growing area or the regions of the growing area and process the normalized or standardized values to classify locations as containing or not containing weeds. As a particular example, the weed classifier 306 may perform k-means clustering to cluster pixels or other image locations into clusters, where at least one of the clusters is then classified as being associated with weed locations and at least one other of the clusters is then classified as associated with non-weed locations. Note that in terms of performing classification, any suitable unsupervised technique (such as clustering) or supervised technique (such as rules-based classification) may be used.

[0044] The classifications or other determinations by the weed classifier 306 may be used in any suitable manner. For example, in some embodiments, the classifications or other determinations by the weed classifier 306 may be provided to a weed map generation function 308, which can process the classifications or other determinations and generate weed maps 310. A weed map generally represents a spatial map of at least one growing area that identifies actual or expected locations of weeds within the growing area(s), possibly along with weed-related information (such as weed type, weed size, weed density, etc.). As a particular example, the weed map generation function 308 may generate a graphical image representing each field or other growing area, where any locations of weeds in the growing area are identified in the graphical image.

[0045] One or more post-processing functions 312 may be used to process the weed maps 310 or the classifications or other determinations generated by the weed classifier 306. The specific post-processing function(s) 312 performed here can vary based on the specific use case. For example, one or more post-processing functions 312 may process the weed maps 310 or other information generated by the architecture 300 to determine which portions of each growing area should be treated with a chemical herbicide or other treatment. As a particular example, the one or more post-processing functions 312 may be used to determine where one or more tractor-based or other land-based sprayers or one or more drone-based or other aerial-based sprayers should be used to treat weed locations. In some cases, if automated sprayers are used, the one or more post- processing functions 312 may actually schedule or initiate spraying of the weed locations (with or without human approval). One possible benefit here is that a grower does not necessarily know how many acres will need to be sprayed before entering a field or other growing area. Mixing too much herbicide results in waste, and mixing too little herbicide results in wasted time since a refill is needed during treatment. The architecture 300 may be used to obtain a more accurate estimate of the area to be treated, which can be used to more accurately estimate the amount of herbicide to be prepared for use. Note that this type of estimation may occur in various ways. For instance, images over a prolonged period of time (such as up to five years or more) could be analyzed toGECO01-00006 14 identify locations where weeds emerged at any time during that period, and those locations could be targeted for treatment. Images over a prolonged period of time (such as up to five years or more) could be analyzed to identify locations where weeds emerged at a similar time of year or similar point in a growing season, and those locations could be targeted for treatment. Images over a shorter period of time (such as the past two weeks) could be analyzed to identify locations where weeds are currently present, and those locations could be targeted for treatment.

[0046] As another example, one or more post-processing functions 312 may process the weed maps 310 or other information generated by the architecture 300 to supplement or augment determinations made by other systems. For instance, weed locations identified by the architecture 300 could be combined with weed locations identified by one or more tractor-based or other land- based sprayers or surveyors or one or more drone-based or other aerial-based sprayers or surveyors. The one or more post-processing functions 312, the one or more sprayers or surveyors, or other component(s) can use the weed locations identified by the architecture 300 in various ways, such as to verify whether the weed locations identified by the one or more sprayers or surveyors are correct. Among other things, this can help to identify false positives (identified weed locations that do not actually contain weeds) and false negatives (locations not identified as weed locations that actually contain weeds). Among other things, this approach can be useful since sprayers and surveyors often include optical systems with cameras, and those sprayers and surveyors can fail to identify smaller weeds that are below their cameras’ thresholds. The architecture 300 may process images within a multi-year time window to estimate where weeds may appear during a current growing season. As a result, one or more sprayers may be used to treat both (i) weeds detected by the sprayer(s) and (ii) areas determined by the architecture 300 to be likely sites of weed emergence during the current growing season.

[0047] As yet another example, one or more post-processing functions 312 may process the weed maps 310 or other information generated by the architecture 300 to estimate the locations of weed seedbanks within each growing area. A single image or even a few images of a growing area will often be inadequate to identify weed seedbanks in that growing area. This is because it is difficult to identify weed seedbanks from a single image or a few images, particularly since the image(s) may be captured at suboptimal time(s) to detect weeds, such as when the image or images are captured at times when a crop canopy is too thick to identify weeds or at times when weeds may not have emerged yet. The ability of the architecture 300 to obtain images (such as satellite images) over a prolonged period of time (such as up to five years or more) allows the one or more post-processing functions 312 to estimate where weeds are likely to appear in the future. TheseGECO01-00006 15 locations can represent estimated seedbank locations, and the estimated seedbank locations can be used to try and reduce or prevent future weed emergence. For instance, a chemical herbicide or other treatment may be scheduled and applied to the location of an estimated seedbank in order to try and prevent weeds from growing at that location in the future.

[0048] As still another example, one or more post-processing functions 312 may process the weed maps 310 or other information generated by the architecture 300 to estimate whether weeds in a particular growing area are showing signs of herbicide resistance. For instance, this may involve the use of two periods of time, one before an herbicide application and one after the herbicide application. As a particular example, images captured during a two-week period or other period before the herbicide application may be analyzed to identify weed locations. Images captured during a two-week period or other period after the herbicide application may also be analyzed to identify weed locations. Note that the images captured after the herbicide application need not represent images captured immediately after the herbicide application but may instead represent or include images captured after some delay (such as one week) that allows time for the herbicide to function. A comparison of the weed locations before and after the herbicide application can be performed, and a spatial analysis can be performed to identify any weeds that appear to have resisted the herbicide application. These weeds can be marked as being a high priority for growers or agronomists to scout or for subsequent application of a different herbicide. This is because it is often highly desirable to kill weeds that are showing signs of herbicide resistance quickly so the weeds do not spread. It is also possible to use this type of approach to identify weeds that may be herbicide resistant during resistance testing being carried out on weeds in the growing area. Additional information associated with identifying herbicide resistance may be found in International Patent Publication No. WO 2023 / 131851 (which is hereby incorporated by reference in its entirety).

[0049] In one or more of these examples, the one or more post-processing functions 312 may be used to predict (based on prior locations of weeds) likely future locations of weeds. That is, the one or more post-processing functions 312 may predict where weeds are likely to grow in the future based on locations where weeds have grown in the past or are currently growing. For instance, the one or more post-processing functions 312 may predict how weeds that are currently growing are likely spreading, and an herbicide application or other treatment may be initiated for the current locations of the weeds and locations where the weeds are likely spreading.

[0050] If weed maps 310 or other results generated by the architecture 300 are used to recommend or trigger actions, the recommended or triggered actions may include various forms ofGECO01-00006 16 treatments. For example, an herbicide application may be recommended or triggered, which generally involves spraying or other application of an herbicide once. A multi-rate herbicide application may be recommended or triggered, which generally involves multiple applications of herbicide at different rates at different times. A multi-herbicide application may be recommended or triggered, which generally involves multiple applications of different herbicides (possibly at different rates) at different times. An increase in seeding density may be recommended or triggered, which generally involves planting or otherwise increasing the density of crop seeds in areas where more weeds are growing (such as in an attempt to crowd out the weeds). A change in crop may be recommended or triggered, which generally involves planting or otherwise placing a different crop in areas where more weeds are growing (such as in an attempt to crowd out the weeds). A blanket application of herbicide may be recommended or triggered if numerous weed clusters covering a large portion of a growing area are identified, which generally involves applying herbicide over most or all of the growing area. A targeted application of nutrients and / or fertilizer may be recommended or triggered, which generally involves application of one or more nutrients and / or fertilizer to an area to help promote crop growth (which may crowd out weeds).

[0051] Although FIGURE 3 illustrates one example of an architecture 300 supporting the estimation of weed population dynamics in a field or other growing area, various changes may be made to FIGURE 3. For example, various components or functions in FIGURE 3 may be combined, further subdivided, replicated, omitted, or rearranged and additional components or functions may be added according to particular needs.

[0052] One issue that can affect fields and other growing areas is the fact that weeds can germinate and grow in different parts of the growing areas at different times. FIGURE 4 illustrates an example aggregation of weed locations to support the estimation of weed population dynamics in a field or other growing area according to this disclosure. For ease of explanation, the aggregation of weed locations shown in FIGURE 4 is described as being performed using the application server 106 in the system 100 shown in FIGURE 1, where the application server 106 may be implemented using one or more instances of the device 200 shown in FIGURE 2 and may implement at least part of the architecture 300 shown in FIGURE 3. However, aggregation of weed locations may be performed using any other suitable device(s) with any other suitable architecture(s) and in any other suitable system(s).

[0053] As shown in FIGURE 4, a weed map 402 is associated with a growing area during spring of a current growing season, a weed map 404 is associated with the growing area during early summer of the current growing season, and a weed map 406 is associated with the growingGECO01-00006 17 area during late summer of the current growing season. Indicators 408a-408b in the weed maps 402-406 are used to identify weed locations, where indicators 408a are associated with younger weeds and indicators 408b are associated with older weeds.

[0054] As can be seen in the example of FIGURE 4, a weed map 410 associated with the growing area for the entire season could identify all of the weed locations experienced in the growing area over the entire growing season. However, the weed map 410 typically could not be generated using a single image of the growing area during one of the seasons since that image would not provide an accurate picture of the complete underlying seedbank in the growing area. Because the architecture 300 is able to obtain images, such as from one or more satellites, over a longer period of time, the architecture 300 is able to generate the weed map 410 to accurately identify weed locations over the growing season.

[0055] Another issue that can affect fields and other growing areas is the fact that weed patterns can change from one growing season to the next. FIGURE 5 illustrates example historical data used for estimation of weed population dynamics in a field or other growing area according to this disclosure. For ease of explanation, estimation based on historical data shown in FIGURE 5 is described as being performed using the application server 106 in the system 100 shown in FIGURE 1, where the application server 106 may be implemented using one or more instances of the device 200 shown in FIGURE 2 and may implement at least part of the architecture 300 shown in FIGURE 3. However, estimation based on historical data may be performed using any other suitable device(s) with any other suitable architecture(s) and in any other suitable system(s).

[0056] As shown in FIGURE 5, weed maps 502-510 are associated with different growing seasons for a single growing area, and the different growing seasons have different weed distributions within the growing area. The differences in weed distributions here can be due to a number of factors, such as weather, soil, topography, crop competition, herbicides used, and crop management techniques. As a particular example, different crops may be planted during different growing seasons, and some crops may be better than other crops when competing with weeds.

[0057] In order to predict the full and dynamically-changing scope of actual and potential weed emergence, weed patterns over multiple growing seasons may be used. However, this type of information is generally not available when only a few images of the growing area are captured at random times. Because the architecture 300 is able to obtain images, such as from one or more satellites, over a prolonged period of time (possibly multiple years), the architecture 300 is able to predict weed emergence patterns more accurately.

[0058] Although FIGURE 4 illustrates one example of aggregation of weed locations toGECO01-00006 18 support the estimation of weed population dynamics in a field or other growing area and FIGURE 5 illustrates one example of historical data used for estimation of weed population dynamics in a field or other growing area, various changes may be made to FIGURES 4 and 5. For example, while the growing area shown in FIGURES 4 and 5 is generally rectangular, each growing area can have any suitable size, shape, and dimensions. Also, the weed maps or other weed distributions shown in FIGURES 4 and 5 are examples only and can easily vary depending on the circumstances. In addition, while FIGURE 5 shows five years of historical data, the architecture 300 may be configured to use historical data over any other suitable time period.

[0059] The following describes example processes that may be used to estimate weed population dynamics in fields or other growing areas. For example, these processes may be used by various functions of the architecture 300. In the following discussion, it is often assumed that lower-resolution satellite images form at least part of the data being analyzed during these processes. Lower-resolution satellite images are available for many growing areas at a relatively high frequency, such as when two or more satellite images of a growing area are captured each week and made available to the architecture 300. These satellite images can therefore be used to estimate locations where weeds have appeared in a field at one or more times over a time window of interest (such as one or more years, one or more seasons, or one or more months). “Lower- resolution” is used here since satellite imagery is generally of lower resolution than other options, such as images captured using drones, tractors, or other systems. Higher-resolution imagery can also be used in these processes if the higher-resolution imagery exists at the desired frequency. It is also possible for infrequent higher-resolution imagery to be incorporated into the processes simply as additional data that can be processed along with lower-resolution imagery.

[0060] FIGURE 6 illustrates an example method 600 for estimating weed population dynamics in a field or other growing area according to this disclosure. For ease of explanation, the method 600 shown in FIGURE 6 is described as being implemented using the application server 106 in the system 100 shown in FIGURE 1, where the application server 106 may be implemented using one or more instances of the device 200 shown in FIGURE 2 and may implement at least part of the architecture 300 shown in FIGURE 3. However, the method 600 may be implemented using any other suitable device(s) and in any other suitable system(s), and the method 600 may be used with any other suitable architecture(s).

[0061] As shown in FIGURE 6, images and optionally other data associated with one or more growing areas within a specified time window are obtained at step 602. This could include, for example, the at least one processing device 202 of the application server 106 obtaining satelliteGECO01-00006 19 images and optionally other images or other data from one or more data sources 302. In some cases, the images may include lower-resolution satellite images of at least one field or other growing area as captured by one or more satellites. Depending on the circumstances, the at least one processing device 202 of the application server 106 may obtain additional information, such as images captured using drones, tractors, or other systems; human-collected scouting data; meteorological data; soil type data; weed species data; data associated with prior herbicide applications; management practices data; or any suitable combination thereof. Specific examples of additional data can include soil properties of the growing area, elevation / topography of the growing area, presence of water in the growing area, or any combination thereof.

[0062] Information associated with individual regions in the growing area(s) and information associated with the overall growing area(s) are analyzed at step 604. This could include, for example, the at least one processing device 202 of the application server 106 performing the one or more data processing functions 304 in order to associate the image data and other obtained data with specific regions of the growing area(s). For instance, image data and optionally other data can be separated or otherwise associated with regions of a growing area arranged in a grid pattern. This could also include the at least one processing device 202 of the application server 106 performing the one or more data processing functions 304 in order to identify one or more characteristics of each region of each growing area and one or more characteristics of the overall growing area. As a particular example, the at least one processing device 202 of the application server 106 may identify, for each growing area, an average NDVI, NDRE, RGB, NIR, or other value(s) or any combination thereof for each region of that growing area and an average NDVI value or other value(s) or any combination thereof for that growing area. These values may be generated across different dates within the specified time window, such as when the average values are determined for each day within the specified time window for which at least one satellite image or other imagery is available.

[0063] Regions of the growing area(s) are classified as containing weeds or not containing weeds at step 606. This could include, for example, the at least one processing device 202 of the application server 106 using the weed classifier 306 to analyze the average values or other data produced by the one or more data processing functions 304 to estimate whether each region of each growing area is likely or not likely to contain weeds. As a particular example, the weed classifier 306 may be used to select a subset of images from the specified time window that adequately show the growing area and, for those images, analyze the average values or other data produced by the one or more data processing functions 304 to estimate whether each region of each growing area isGECO01-00006 20 likely or not likely to contain weeds. In some embodiments, the weed classifier 306 may perform clustering or other classification to identify which pixels in the subset of images are or are not likely associated with weeds. In some cases, the classification may be fully automated. In other cases, the classification may occur in a semi-supervised manner in which classifications by the weed classifier 306 are presented to one or more human personnel for review and approval, rejection, or modification.

[0064] The classification results may be used in any suitable manner. For example, one or more weed maps may be generated for each growing area based on the classification results at step 608. This could include, for example, the at least one processing device 202 of the application server 106 performing the weed map generation function 308 to generate at least one weed map 310 for each growing area. In some cases, the weed map generation function 308 can apply spatial clustering or spatial filtering to the pixels that are identified as being associated with weeds. The spatial clustering or spatial filtering can create smoother boundaries (rather than pixelated boundaries) for locations associated with weeds.

[0065] The one or more weed maps (or the classification results) may be used to perform one or more functions related to the growing area(s) at step 610. This could include, for example, the at least one processing device 202 of the application server 106 performing the one or more post- processing functions 312 to predict likely locations of weeds in the future, identify where one or more herbicide applications or other treatments should occur, augment weed determinations by land-based or aerial-based sprayers or surveyors, estimate locations of weed seedbanks, or identify instances of possible herbicide resistance.

[0066] Although FIGURE 6 illustrates one example of a method 600 for estimating weed population dynamics in a field or other growing area, various changes may be made to FIGURE 6. For example, while shown as a series of step, various steps in FIGURE 6 may overlap, occur in parallel, occur in a different order, or occur any number of times (including zero times). As a particular example, FIGURE 6 assumes that all data is collected and split or otherwise associated with different regions of a growing area prior to processing. Thus, for instance, if high-resolution drone data becomes available for a specific day, that data can be processed along with satellite imagery, and the results can be used for weed classification. However, in other embodiments, data can be combined with the results of the classification. As a particular example, a map of detected weeds in a growing area generated using an optical spot sprayer on a specific day can be combined with a weed map 310 for that growing area (generated after weed classification), and the resulting combination can form a map of estimated weeds in the growing area.GECO01-00006 21

[0067] FIGURE 7 illustrates an example method 700 for obtaining information to be used for estimating weed population dynamics in a field or other growing area according to this disclosure. The method 700 may, for example, represent a process that can be performed during or as part of step 602 in the method 600 shown in FIGURE 6. For ease of explanation, the method 700 shown in FIGURE 7 is described as being implemented using the application server 106 in the system 100 shown in FIGURE 1, where the application server 106 may be implemented using one or more instances of the device 200 shown in FIGURE 2 and may implement at least part of the architecture 300 shown in FIGURE 3. However, the method 700 may be implemented using any other suitable device(s) and in any other suitable system(s), and the method 700 may be used with any other suitable architecture(s).

[0068] As shown in FIGURE 7, satellite images of at least one growing area are obtained at step 702. This could include, for example, the at least one processing device 202 of the application server 106 obtaining satellite images from one or more data sources 302. The satellite images can span a specified time window. In some embodiments, the specified time window can be based on when a user wants an estimate of germinated weeds to be generated by the architecture 300. As a particular example, the specified time window may encompass all satellite images of the growing area(s) captured over the past five years or other prolonged period of time. As another particular example, the user may wish to compare weeds at one time period (such as one part of a growing season) to weeds at another time period (such as another part of the same growing season or an earlier growing season), and the specified time window may encompass all satellite images of the growing area(s) captured for those two time periods. In some embodiments, the satellite images obtained here may be lower-resolution images. In other embodiments, at least some of the satellite images could represent higher-resolution images, such as if one or more satellites capable of capturing higher-resolution images are available.

[0069] Additional data associated with the growing area(s) may optionally be obtained at step 704. This could include, for example, the at least one processing device 202 of the application server 106 obtaining additional images or other data from one or more data sources 302. For instance, the additional data may include images captured using drones, tractors, or other systems; human-collected scouting data; meteorological data; soil type data; weed species data; data associated with prior herbicide applications; management practices data; or any suitable combination thereof. As noted above, specific examples of additional data could include soil properties of the growing area, elevation / topography of the growing area, presence of water in the growing area, or any combination thereof. Note, however, that the method 700 may not require useGECO01-00006 22 of any additional data.

[0070] The obtained data is filtered at step 706. This could include, for example, the at least one processing device 202 of the application server 106 identifying images where part or all of a growing area is obscured, such as by cloud cover or by snow. Images in which part or all of the growing area is obscured can be removed from further processing. The filtered images are processed to generate initial information about the growing area(s) at step 708. This could include, for example, the at least one processing device 202 of the application server 106 processing the filtered images to identify or create multiple types of information. As particular examples, the at least one processing device 202 of the application server 106 could convert each filtered image into RGB and NIR bands, generate NDVI and NDRE index values, and / or generate data for other multi- spectral or hyper-spectral bands.

[0071] Although FIGURE 7 illustrates one example of a method 700 for obtaining information to be used for estimating weed population dynamics in a field or other growing area, various changes may be made to FIGURE 7. For example, while shown as a series of step, various steps in FIGURE 7 may overlap, occur in parallel, occur in a different order, or occur any number of times (including zero times).

[0072] FIGURE 8 illustrates an example method 800 for processing information for estimating weed population dynamics in a field or other growing area according to this disclosure. The method 800 may, for example, represent a process that can be performed during or as part of step 604 in the method 600 shown in FIGURE 6. For ease of explanation, the method 800 shown in FIGURE 8 is described as being implemented using the application server 106 in the system 100 shown in FIGURE 1, where the application server 106 may be implemented using one or more instances of the device 200 shown in FIGURE 2 and may implement at least part of the architecture 300 shown in FIGURE 3. However, the method 800 may be implemented using any other suitable device(s) and in any other suitable system(s), and the method 800 may be used with any other suitable architecture(s).

[0073] As shown in FIGURE 8, the images are processed to identify one or more characteristics for each overall growing area over time at step 802. This could include, for example, the at least one processing device 202 of the application server 106 processing the data associated with each growing area in order to identify one or more characteristics for that growing area. Any suitable characteristic(s) can be identified here for each of the growing areas. As a particular example, the at least one processing device 202 of the application server 106 may process the RGB data, NIR data, NDVI values, NDRE index values, other multi-spectral or hyper-spectral data, orGECO01-00006 23 any combination thereof generated during the method 700. In some cases, the at least one processing device 202 of the application server 106 may process this information or other information to generate average red, green, blue, NDVI, NDRE, and / or NIR values (including any combination thereof) for each day at least one satellite image is available within the specified time window. Also, in some cases, the average values may be normalized or standardized across the growing areas.

[0074] In some embodiments, the one or more characteristics for each overall growing area may be determined by identifying one or more characteristics for different regions of a growing area and combining the one or more characteristics for the different regions to identify the one or more characteristics for the growing area. One example of this is shown in FIGURE 9, where an image 900 of a growing area is overlaid with grid lines 902 that divide the growing area into a number of individual regions 904. Indicators 906 identify portions of the growing area where weeds are or might be present, and it can be seen that these portions of the growing area partially or completely cover one or more of the individual regions 904. One or more characteristics for each region 904 may be determined, and the one or more characteristics may be averaged across all of the regions 904 or otherwise processed to identify the one or more characteristics for the overall growing area. Again, this may be done for each day at least one satellite image is available within the specified time window, and the average values may be normalized or standardized across the growing areas in some cases. Note, however, that this is not necessarily required, and the one or more characteristics for the overall growing area may be determined in any suitable manner with or without dividing the growing area into regions 904.

[0075] The one or more characteristics for each overall growing area are compared to one or more desired values or ranges at step 804, and a subset of the images that fall within the specified time window is selected based on the comparison(s) at 806. This could include, for example, the at least one processing device 202 of the application server 106 selecting images having one or more overall characteristics within one or more desired ranges. As a particular example, the at least one processing device 202 of the application server 106 could select images within the specified time window having an average NDVI value across the entire growing area greater than 0.1 and less than 0.25. This type of condition may indicate that the associated images are likely to capture actual germinated weeds without excessive crop interference. Of course, any other or additional condition(s) or data may be used here to select the subset of images. For instance, in addition to using average NDVI values, another approach may consider local temperature, local precipitation, the type of crop planted, the crop seeding date, the crop harvest date, the herbicide applicationGECO01-00006 24 records, or any combination thereof when selecting the subset of images. As a particular example of this, assume multiple images are available each week for a given year. It is possible to use a combination of local temperature, precipitation, crop information, herbicide application records, and NDVI patterns based on satellite images (such as the rise and fall of the average NDVI representing crop stage in a growing area) to decide that only satellite images for a certain time period earlier in a season (such as May 1 through June 15) and a certain time period later in the season (such as August 15 to October 1) will be used in the classifying process to detect weeds. These time periods may be selected as encompassing times when weeds were likely growing and visible before a crop grew and after the crop has been harvested.

[0076] Although FIGURE 8 illustrates one example of a method 800 for processing information for estimating weed population dynamics in a field or other growing area, various changes may be made to FIGURE 8. For example, while shown as a series of steps, various steps in FIGURE 8 may overlap, occur in parallel, occur in a different order, or occur any number of times (including zero times). Although FIGURE 9 illustrates one example of a grid-based approach for processing information for estimating weed population dynamics in a field or other growing area, various changes may be made to FIGURE 9. For instance, while the growing area shown in FIGURE 9 is generally rectangular, each growing area can have any suitable size, shape, and dimensions. Also, while the grid pattern shown in FIGURE 9 uses horizontal and vertical grid lines 902 that are evenly spaced, the grid pattern may be defined in any other suitable uniform or non- uniform manner.

[0077] FIGURE 10 illustrates an example method 1000 for classification to support estimation of weed population dynamics in a field or other growing area according to this disclosure. The method 1000 may, for example, represent a process that can be performed during or as part of steps 604 and 606 in the method 600 shown in FIGURE 6. For ease of explanation, the method 1000 shown in FIGURE 10 is described as being implemented using the application server 106 in the system 100 shown in FIGURE 1, where the application server 106 may be implemented using one or more instances of the device 200 shown in FIGURE 2 and may implement at least part of the architecture 300 shown in FIGURE 3. However, the method 1000 may be implemented using any other suitable device(s) and in any other suitable system(s), and the method 1000 may be used with any other suitable architecture(s).

[0078] As shown in FIGURE 10, a grid is defined across each growing area at step 1002. This could include, for example, the at least one processing device 202 of the application server 106 defining a grid using horizontal and vertical grid lines 902 as shown in FIGURE 9. Note that if theGECO01-00006 25 one or more characteristics for each overall growing area are determined in step 802 by identifying one or more characteristics for different regions 904 of a growing area and combining the one or more characteristics for the different regions, the grid defined in step 1002 may or may not represent the same grid used in step 802.

[0079] The images in the selected subset are processed to identify one or more characteristics for each grid cell of each growing area over time at step 1004. This could include, for example, the at least one processing device 202 of the application server 106 processing the data associated with each region 904 in order to identify one or more characteristics for that region 904. Any suitable characteristic(s) can be identified here for each of the regions 904. As a particular example, the at least one processing device 202 of the application server 106 may process the RGB data, NIR data, NDVI values, NDRE index values, other multi-spectral or hyper-spectral data, or any combination thereof generated during the method 700. In some cases, the at least one processing device 202 of the application server 106 may process this information or other information to generate average red, green, blue, NDVI, NDRE, and / or NIR values (including any combination thereof) for each day at least one satellite image is available within the specified time window. Also, in some cases, these values may be generated for each region 904 and possibly for each pixel within each region 904. Further, in some cases, the average values may be normalized or standardized across each growing area or across all growing areas.

[0080] A machine learning-based classifier or other classifier is applied to the determined characteristic(s) of the regions of the growing area(s) based on the subset of images at step 1006. This could include, for example, the at least one processing device 202 of the application server 106 applying the weed classifier 306 to the average values determined for the regions 904 of each growing area. This can be done for each image within the subset of images, meaning the classification can occur for image data over time. In some embodiments, the weed classifier 306 can be implemented using a clustering or other classification algorithm applied to all pixels of each image within the selected subset of images. The weed classifier 306 here can classify image pixels as belonging to different classes based on one or more characteristics, such as normalized or standardized NDVI, NDRE, RGB, and / or NIR values. The different classes can be assigned as containing weeds or not containing weeds based on properties of that class (such as the NDVI, NDRE, RGB, and / or NIR values). This can be repeated for each image within the selected subset of images over time, and the results can be combined across the specified time window to provide a picture of all weeds that flushed in or grew over that specified time window.

[0081] Also or alternatively, one or more characteristics of the regions of the growing area(s)GECO01-00006 26 are summarized based on the subset of images at step 1008, and a machine learning-based classifier or other classifier is applied to the summarized characteristic(s) of the regions of the growing area(s) at step 1010. This could include, for example, the at least one processing device 202 of the application server 106 summarizing the average values determined for the regions 904 of each growing area and applying the weed classifier 306 to the summarized values. This can be done across all images within the subset of images, meaning the average values for image data over time can be summarized. The one or more characteristics can be summarized in any suitable manner, such as by adding the values of the one or more characteristics across all of the images within the subset of images. In some embodiments, the weed classifier 306 can be implemented using a clustering or other classification algorithm applied to the summarized data for all pixels within the selected subset of images. The weed classifier 306 here can classify image pixels as belonging to different classes based on one or more characteristics, such as summarized NDVI, NDRE, RGB, and / or NIR values. The different classes can be assigned as containing weeds or not containing weeds based on properties of that class (such as the summarized NDVI, NDRE, RGB, and / or NIR values). Note that the machine learning-based classifier or other classifier used during step 1006 and the machine learning-based classifier or other classifier used during step 1010 may be the same classifier or different classifiers.

[0082] Final classifications for the growing area(s) are generated at step 1012. This could include, for example, the at least one processing device 202 of the application server 106 combining the classifications made during step 1006 with the classifications made during step 1010. Any suitable technique can be used here to combine different sets of classifications. For example, the at least one processing device 202 of the application server 106 may perform a union of the classifications made during step 1006 with the classifications made during step 1010, a weighted combination of the classifications made during step 1006 with the classifications made during step 1010, or other suitable combination.

[0083] The final classifications can be stored, output, or used at step 1014. This could include, for example, the at least one processing device 202 of the application server 106 using the final classifications to produce one or more weed maps 310. The weed maps 310 or the final classifications may be processed using the one or more post-processing functions 312 to provide any additional functionality as needed or desired. Examples of this additional functionality are provided above, and the additional functionality can easily vary depending on the specific use case.

[0084] Although FIGURE 10 illustrates one example of a method 1000 for classification to support estimation of weed population dynamics in a field or other growing area, various changesGECO01-00006 27 may be made to FIGURE 10. For example, while shown as a series of step, various steps in FIGURE 10 may overlap, occur in parallel, occur in a different order, or occur any number of times (including zero times). As a particular example, the specific order of the processing can vary, such as when steps 1008-1010 occur before or in parallel with step 1006.

[0085] Using the approaches described above, the one or more characteristics of the regions 904 within each growing area can be used to identify those regions 904 within each growing area that differ from other regions 904 in a manner indicative of weed presence. For example, during times when crops are just starting to grow or after harvesting, the presence of numerous green or other-colored plants can be indicative of the presence of weeds. Thus, differences between the one or more characteristics over time can be used to identify the presence of weeds. Moreover, the one or more characteristics of each overall growing area can be used to identify one or more trends for the overall growing area. For instance, average NDVI values or other values for an entire growing area over time can be used to estimate overall vegetative trends over time in the growing area, such as by allowing the architecture 300 to identify or estimate the current stage of crop growth within the growing area. Collectively, this allows the architecture 300 to understand weed emergence and crop dynamics, as well as to estimate at what point weeds become visible to satellites or other systems that capture the images being processed.

[0086] Note that while the functionality above often assumes that a determination is being made whether weeds are present or likely to be present within one or more growing areas, other or additional weed-related determinations may be made here. For example, instead of or in addition to determining the presence / absence of weeds in one or more growing areas, the described techniques may be used to detect different types of weeds or some measure of the number of weeds in the one or more growing areas. As particular examples, the described techniques may be used to differentiate between grass and broadleaf weeds, differentiate between different taxonomic groups of weeds (such as by detecting different weed species, genera, families, or orders), differentiate between herbicide-resistant weeds and herbicide-susceptible weeds, or differentiate between weeds of different ages. As another example, the described techniques may be used to estimate absolute or relative weed density (sometimes referred to as weed pressure) in the one or more growing areas. In these cases, the differentiation may be possible based on differences in the NDVI, NDRE, RGB, NIR, and / or other values associated with the different weeds or plants in the growing area(s).

[0087] Also note that the functions shown in or described with respect to FIGURES 3 through 10 can be implemented in the application server 106, user device 102a-102d, or other device(s) in any suitable manner. For example, in some embodiments, at least some of the functions shown inGECO01-00006 28 or described with respect to FIGURES 3 through 10 can be implemented or supported using one or more software applications or other software instructions that are executed by the at least one processing device 202 of the application server 106, user device 102a-102d, or other device(s). In other embodiments, at least some of the functions shown in or described with respect to FIGURES 3 through 10 can be implemented or supported using dedicated hardware components. In general, the functions shown in or described with respect to FIGURES 3 through 10 can be performed using any suitable hardware or any suitable combination of hardware and software / firmware instructions. Also, the functions shown in or described with respect to FIGURES 3 through 10 can be performed by a single device or by multiple devices.

[0088] In some embodiments, various functions described in this patent document are implemented or supported by a computer program that is formed from computer readable program code and that is embodied in a computer readable medium. The phrase “computer readable program code” includes any type of computer code, including source code, object code, and executable code. The phrase “computer readable medium” includes any type of medium capable of being accessed by a computer, such as read only memory (ROM), random access memory (RAM), a hard disk drive (HDD), a compact disc (CD), a digital video disc (DVD), or any other type of memory. A “non-transitory” computer readable medium excludes wired, wireless, optical, or other communication links that transport transitory electrical or other signals. A non-transitory computer readable medium includes media where data can be permanently stored and media where data can be stored and later overwritten, such as a rewritable optical disc or an erasable storage device.

[0089] It may be advantageous to set forth definitions of certain words and phrases used throughout this patent document. The terms “application” and “program” refer to one or more computer programs, software components, sets of instructions, procedures, functions, objects, classes, instances, related data, or a portion thereof adapted for implementation in a suitable computer code (including source code, object code, or executable code). The term “communicate,” as well as derivatives thereof, encompasses both direct and indirect communication. The terms “include” and “comprise,” as well as derivatives thereof, mean inclusion without limitation. The term “or” is inclusive, meaning and / or. The phrase “associated with,” as well as derivatives thereof, may mean to include, be included within, interconnect with, contain, be contained within, connect to or with, couple to or with, be communicable with, cooperate with, interleave, juxtapose, be proximate to, be bound to or with, have, have a property of, have a relationship to or with, or the like. The phrase “at least one of,” when used with a list of items, means that different combinations of one or more of the listed items may be used, and only one item in the list may be needed. ForGECO01-00006 29 example, “at least one of: A, B, and C” includes any of the following combinations: A, B, C, A and B, A and C, B and C, and A and B and C.

[0090] The description in the present application should not be read as implying that any particular element, step, or function is an essential or critical element that must be included in the claim scope. The scope of patented subject matter is defined only by the allowed claims. Moreover, none of the claims invokes 35 U.S.C. § 112(f) with respect to any of the appended claims or claim elements unless the exact words “means for” or “step for” are explicitly used in the particular claim, followed by a participle phrase identifying a function. Use of terms such as (but not limited to) “mechanism,” “module,” “device,” “unit,” “component,” “element,” “member,” “apparatus,” “machine,” “system,” “processor,” or “controller” within a claim is understood and intended to refer to structures known to those skilled in the relevant art, as further modified or enhanced by the features of the claims themselves, and is not intended to invoke 35 U.S.C. § 112(f).

[0091] While this disclosure has described certain embodiments and generally associated methods, alterations and permutations of these embodiments and methods will be apparent to those skilled in the art. Accordingly, the above description of example embodiments does not define or constrain this disclosure. Other changes, substitutions, and alterations are also possible without departing from the spirit and scope of this disclosure, as defined by the following claims.

Claims

GECO01-00006 30 WHAT IS CLAIMED IS:

1. A method comprising: obtaining satellite images of a growing area, the satellite images captured repeatedly within a specified time window; selecting at least some of the satellite images for processing; identifying one or more overall characteristics of the growing area at one or more times using at least some of the selected satellite images; selecting a subset of the selected satellite images within the specified time window based on the one or more overall characteristics; dividing the growing area into multiple regions; identifying one or more characteristics of each of the regions of the growing area based on the subset of the selected satellite images; and using at least one classifier to classify each of the regions of the growing area as either containing weeds or not containing weeds based on the one or more characteristics of the regions of the growing area and the one or more overall characteristics of the growing area.

2. The method of Claim 1, wherein: each of the satellite images has a specified resolution; the method further comprises obtaining additional images having a resolution higher than the specified resolution; and using the classifier to classify each of the regions of the growing area is further based on the additional images.

3. The method of Claim 2, wherein the additional images are obtained from at least one of: one or more higher-resolution satellites, one or more land-based optical sprayers or surveyors, or one or more aerial-based optical sprayers or surveyors.

4. The method of any of Claims 1-3, wherein identifying the one or more characteristics of each of the regions of the growing area comprises: identifying the one or more characteristics of each of the regions of the growing area for each of the satellite images in the subset of the selected satellite images.

5. The method of any of Claims 1-4, wherein the one or more characteristics of each of the regions of the growing area comprise at least one of:GECO01-00006 31 an average normalized difference vegetation index (NDVI) value based on image data associated with the region in each of the satellite images in the subset of the selected satellite images; an average normalized difference red edge (NDRE) index value based on the image data associated with the region in each of the satellite images in the subset of the selected satellite images; an average red value based on the image data associated with the region in each of the satellite images in the subset of the selected satellite images; an average green value based on the image data associated with the region in each of the satellite images in the subset of the selected satellite images; an average blue value based on the image data associated with the region in each of the satellite images in the subset of the selected satellite images; and an average near infrared (NIR) value based on the image data associated with the region in each of the satellite images in the subset of the selected satellite images.

6. The method of any of Claims 1-5, wherein identifying the one or more overall characteristics of the growing area comprises: identifying the one or more overall characteristics of the growing area for each of the selected satellite images.

7. The method of any of Claims 1-6, wherein the one or more overall characteristics of the growing area comprises an average NDVI value based on image data associated with the growing area in each of the selected satellite images.

8. The method of any of Claims 1-7, wherein selecting the subset of the selected satellite images within the specified time window comprises: selecting the satellite images within the specified time window having an average NDVI value within a specified range.

9. The method of any of Claims 1-8, wherein dividing the growing area into the multiple regions comprises dividing the growing area based on a grid pattern.

10. The method of any of Claims 1-9, wherein selecting at least some of the satelliteGECO01-00006 32 images for processing comprises: filtering the satellite images to remove images in which all or a portion of the growing area is obscured.

11. The method of any of Claims 1-10, wherein using the at least one classifier to classify each of the regions of the growing area as either containing weeds or not containing weeds comprises: applying a classifier to the one or more characteristics of each region of the growing area for each pixel of each of the satellite images in the subset of the selected satellite images; and combining classification results across all of the satellite images in the subset of the selected satellite images.

12. The method of any of Claims 1-11, wherein using the at least one classifier to classify each of the regions of the growing area as either containing weeds or not containing weeds comprises: summarizing the one or more characteristics of each region of the growing area across all of the satellite images in the subset of the selected satellite images to generate one or more summarized characteristics of each region of the growing area; and applying a classifier to the one or more summarized characteristics of each region of the growing area for each pixel of each of the satellite images in the subset of the selected satellite images.

13. The method of any of Claims 1-12, wherein: using the at least one classifier to classify each of the regions of the growing area as either containing weeds or not containing weeds is semi-supervised; and classifications generated using the at least one classifier are approved, rejected, or modified by at least one person.

14. The method of any of Claims 1-13, further comprising: generating at least one weed map for the growing area based on classifications generated using the at least one classifier.

15. The method of any of Claims 1-14, wherein the satellite images comprise one orGECO01-00006 33 more satellite images captured per week within the specified time window.

16. The method of any of Claims 1-15, wherein the specified time window spans one or a portion of a growing season of a crop planted in the growing area.

17. The method of any of Claim 16, further comprising: combining multiple weed maps for the growing area, the weed maps associated with multiple growing seasons of one or more crops planted in the growing area.

18. The method of any of Claims 1-17, further comprising: predicting future locations of weeds in the growing area based on classifications generated using the at least one classifier.

19. The method of any of Claims 1-18, further comprising: estimating at least one location of at least one weed seedbank in the growing area based on classifications generated using the at least one classifier.

20. The method of any of Claims 1-19, further comprising: detecting whether any weeds in the growing area have or might be developing an herbicide resistance based on classifications generated using the at least one classifier.

21. The method of any of Claims 1-20, further comprising: identifying one or more portions of the growing area to receive one or more weed treatments based on classifications generated using the at least one classifier.

22. The method of any of Claims 1-21, further comprising: augmenting an identification of one or more portions of the growing area to receive one or more weed treatments based on classifications generated using the at least one classifier.

23. The method of Claim 22, wherein augmenting the identification comprises augmenting weed identification by a system that uses a proximal camera to identify weeds, the classifications generated using the at least one classifier assisting in the weed identification for weeds that are below a threshold of the proximal camera.GECO01-00006 34 24. The method of any of Claims 1-23, further comprising: receiving additional data associated with the growing area; and processing the additional data associated with at least some of the regions of the growing area; wherein using the classifier to classify each of the regions of the growing area is further based on at least some of the additional data.

25. The method of Claim 24, wherein the additional data comprises at least one of: one or more soil properties of the growing area; an elevation or topography of the growing area; or a presence of water in the growing area.

26. The method of any of Claims 1-25, wherein selecting the subset of the selected satellite images is based on at least one of: local temperature in the growing area; local precipitation in the growing area; a type of crop planted in the growing area; a seeding date of the crop planted in the growing area; a harvest date of the crop planted in the growing area; or one or more records of one or more herbicide applications in the growing area.

27. The method of any of Claims 1-26, further comprising: detecting different types, densities, or numbers of weeds in the growing area.

28. An apparatus comprising: at least one processing device configured to perform the method of any of Claims 1-27.

29. A non-transitory computer readable medium storing computer readable program code that when executed causes at least one processor to perform the method of any of Claims 1- 27.

Images