Mapping field anomalies using digital images and machine learning models

Abstract

A computer-implemented method for generating an enhanced field anomaly map using digital imagery and machine learning models is disclosed; in one embodiment, the method comprises: obtaining a Shapefile that defines boundaries of an agricultural parcel and boundaries of the field containing the parcel; obtaining a plurality of parcel images; calibrating and pre-processing the plurality of parcel images to create a parcel map of the agricultural parcel at a parcel level; based on the parcel map, generating a parcel grid; based on the grid and the grid map, generating a plurality of parcel tiles; based on the plurality of parcel tiles, generating, using a machine learning model and a plurality of image classifiers corresponding to one or more anomalies, a set of classified parcel images that exhibits at least one anomaly;Based on the set of classified parcel images, generate a parcel anomaly map for the agricultural parcel.

Description

Mapping field anomalies using digital images and machine learning models FIELD OF INVENTION One technical field covered by this disclosure is computer-implemented analysis of digital images. Another technical field is computer-implemented interpretation and analysis of digital images of agricultural fields, typically images obtained above ground using satellites, unmanned aerial vehicles, or other aircraft. BACKGROUND OF THE INVENTION The approaches described in this section are approaches that could be promoted, but are not necessarily approaches that have been previously conceived or promoted. Therefore, unless otherwise stated, any of the approaches described in this section should be assumed to qualify as a prior technique simply by virtue of its inclusion in this section. One of the initiatives in precision agriculture is to accurately measure the percentage of abnormal areas in agricultural fields. Farmers often want to understand the extent and severity of bending and weeds in their fields, as well as the yield impact of these anomalies. Recently, many imaging approaches, particularly UAV-based imaging methods, have been investigated to detect bending and weeds in fields. For example, Chu et al. (2017) assessed bending rates in maize based on leaf color and plant height information measured by UAVs. Huang et al. (2018) applied high-resolution UAV imaging systems to assess weed distributions within a field. However, there is no systematic approach that can accurately and simultaneously detect and classify bending, bare soil, and weeds in an automated manner. Some approaches to equipment bending or damage detection and weed detection have used expensive sensors such as LiDAR and hyperspectral sensors, or sophisticated and time-consuming post-processing such as Motion Surface (Digital Surface Model). The performance of these approaches is often limited, making them difficult to scale for commercial or multi-field operations. Based on the above, improved and efficient computer-implemented methods are needed to determine anomalies in agricultural fields based on digital images. BRIEF DESCRIPTION OF THE INVENTION The appended claims may serve as a brief description of the invention. ncoonn / i ζπζ / ε / υιλι BRIEF DESCRIPTION OF THE FIGURES In the drawings: Figure 1 illustrates an exemplary computer system that is configured to perform the functions described herein, shown in a field environment with another device with which the system can interoperate. Figures 2A and 2B illustrate two views of an exemplary logical organization of instruction sets in main memory when an exemplary mobile application is loaded for execution. Figure 3 illustrates a programmed process through which the agricultural intelligence computer system generates one or more pre-configured agronomic models using agronomic data provided by one or more data sources. Figure 4 is a block diagram illustrating a computer system on which a modality of the invention can be implemented. Figure 5 illustrates an exemplary form of a timeline view for data entry. Figure 6 illustrates an exemplary form of a spreadsheet view for data entry. Figure 7 illustrates an exemplary digital image processing to generate a field anomaly map using machine learning models. Figure 8 illustrates an exemplary processing of aerial and UAV images to generate a field anomaly map using machine learning models. Figure 9 illustrates an exemplary soil image processing to generate a field anomaly map using machine learning models. Figure 10 illustrates an exemplary soil image processing to generate a field anomaly map using machine learning models and an edge TPU. Figure 11 illustrates an exemplary machine learning approach for classifying images to generate a field anomaly map using machine learning models. Figure 12 illustrates an exemplary image classification using a machine learning approach to generate a field anomaly map using machine learning models. Figure 13 illustrates an example of image classification using a machine learning approach to generate a field anomaly map using machine learning models. Figure 14 illustrates an example of a neural network configuration to generate a field anomaly map using machine learning models. Figure 15 illustrates an exemplary flowchart for processing aerial and UAV imagery to generate a field anomaly map using machine learning models. Figure 16 illustrates an exemplary flowchart for processing ground images in order to generate a field anomaly map using machine learning models. DETAILED DESCRIPTION OF THE INVENTION In the following description, for explanatory purposes, numerous specific details are set out to provide a complete understanding of this disclosure. However, it will become apparent that the modalities can be practiced without these specific details. In other cases, well-known structures and devices are shown in block diagram form to avoid unnecessarily obscuring this disclosure. The modalities are disclosed in sections according to the following guideline: 1. General perspective 2. Exemplary agricultural intelligence computer system 2.1. General structural perspective 2.2. Overview of the application program 2.3. Data consumption for the computer system 2.4. General perspective of the process - agronomic model training 2.5. Implementation Example - Hardware Overview 3. Digital Image Processing Approach 3.1 Digital image processing for aerial images 3.2 Digital image processing for soil images 4. Exemplary processing of aerial and UAV images 5. Exemplary soil image processing 6. Exemplary implementation of soil image processing 6.1 Exemplary Edge Computing Implementation 6.2 Exemplary TPU Edge Computing Implementation 7. Exemplary machine learning approach 8. Exemplary classifiers 9. Exemplary image classification 10. Example neural network configuration 11. Example flowchart for aerial and UAV image processing 12. Example flowchart for soil image processing 13. Benefits of some modalities ncoonn / i ζπζ / β / υιλι Overview In one modality, a machine learning approach is provided for the detection and mapping of bending, bare soil, and weed patches in a cornfield using color (Red-Green-Blue) and near-infrared (NIR) imagery collected from an aircraft such as an unmanned aerial vehicle (UAV) platform, and / or ground vehicle platform. Ground vehicles may include harvesters, mowers, or other equipment operating in agricultural fields. Digital imagery used in modality may comprise multichannel data with red pixel, green pixel, blue pixel, and NIR pixel or other components. 2. Exemplary agricultural intelligence computer system 2.1. General structural perspective Figure 1 illustrates an exemplary computer system configured to perform the functions described herein, shown in a field environment with another device with which the system can interoperate. In one embodiment, a user 102 owns, operates, or has a field management computing device 104 at a field location or associated with a field location, such as a field designated for agricultural activities or an administrative location for one or more agricultural fields. The field management computing device 104 is programmed or configured to provide field data 106 to an agricultural intelligence computer system 130 through one or more networks 109. Examples of field data 106 include (a) identification data (e.g., area, field name, field identifiers, geographic identifiers, boundary identifiers, crop identifiers, and any other convenient data that may be used to identify arable land, such as a common land unit (CLU), lot and block number, parcel number, geographic coordinates and boundaries, Farm Serial Number (FSN), farm number, apparatus number, field number, section, municipality, and / or range), (b) harvest data (e.g., crop type, crop variety, crop rotation, whether the crop is being grown organically, harvest date, Actual Production History (APH), expected yield, yield, crop price, crop income, grain moisture, tillage practice, and previous growing season information), (c) soil data (e.g., type, composition, pH,(e.g., organic matter (OM), cation exchange capacity (CEC)), (d) planting data (e.g., planting date, seed type, relative maturity (RM) of planted seeds, seed population), (e) fertilizer data (e.g., nutrient type (Nitrogen, Phosphorus, Potassium), application type, application date, quantity, source, method), (f) chemical application data (e.g., pesticide, herbicide, fungicide, other substance or mixture of substances intended for use as a plant regulator, defoliant, or desiccant, application date, quantity, source, method), (g) irrigation data (e.g., application date, quantity, source, method), (h) climate data (e.g., precipitation, precipitation rate, predicted precipitation, water runoff rate region, temperature, wind, forecast, pressure, visibility, clouds, heat index, dew point, humidity, snow depth, air quality,(sunrise, sunset), (i) image generation data (e.g., light spectrum information and image generation from a sensor of agricultural equipment, camera, computer, smartphone, tablet, unmanned aerial vehicle, aircraft or satellites), (j) exploration observations (photos, videos, free-form notes, voice recordings, voice transcripts, weather conditions (temperature, precipitation (current and over time), soil moisture, crop growth stage, wind speed, relative humidity, dew point, black layer)) and (k) soil, seed, crop phenology, pest and disease report, as well as prediction sources and databases. A data server computer 108 is communicatively coupled to the agricultural intelligence computer system 130 and is programmed or configured to send external data 110 to the agricultural intelligence computer system 130 via networks 109. The external data server computer 108 may be owned or operated by the same person or legal entity as the agricultural intelligence computer system 130, or by a different person or entity such as a government agency, non-governmental organization (NGO), and / or a private data service provider. Examples of external data include weather data, image generation data, soil data, or statistical data related to crop yields, among others. External data 110 may consist of the same type of information as field data 106.In some embodiments, external data 110 is provided by an external data server 108 owned by the same entity that owns and / or operates the agricultural intelligence computer system 130. For example, the agricultural intelligence computer system 130 may include a data server focused exclusively on a type of data that could otherwise be obtained from third-party sources, such as weather data. In some embodiments, an external data server 108 may be incorporated within the system 130. An agricultural apparatus 111 may have one or more fixed remote sensors 112 attached to it. These sensors are communicatively coupled either directly or indirectly through the agricultural apparatus 111 to the agricultural intelligence computer system 130 and are programmed or configured to send sensor data to the agricultural intelligence computer system 130. Examples of agricultural apparatus 111 include tractors, harvesters, mowers, planters, trucks, fertilizer equipment, aerial vehicles including unmanned aerial vehicles, and any other element of physical machinery or hardware, typically mobile machinery, that can be used in tasks associated with agriculture.In some embodiments, a single unit of apparatus 111 may comprise a plurality of sensors 112 that are locally coupled in a network within the apparatus; the Controller Area Network (CAN) is an example of such a network that can be installed in combine harvesters, mowers, sprayers, and cultivators. The application controller 114 is communicatively coupled to the agricultural intelligence computer system 130 via networks 109 and is programmed or configured to receive one or more command sequences that are used to control an operating parameter of an agricultural vehicle or implemented from the agricultural intelligence computer system 130.For example, a controller area network (CAN) bus interface can be used to enable communication from the agricultural intelligence computer system 130 to the agricultural implement 111, such as in the manner used by CLIMATE FIELDVIEW DRIVE, available from The Climate Corporation, San Francisco, California. The sensor data can consist of the same type of information as the field data 106. In some embodiments, remote sensors 112 may not be fixed to an agricultural implement 111 but may be remotely located in the field and can communicate with the network 109. The apparatus 111 may comprise a cockpit computer 115 programmed with a cockpit application, which may comprise a version or variant of the mobile application for the apparatus 104 further described in other sections of this document. In one embodiment, the cockpit computer 115 comprises a compact computer, frequently a tablet-sized computer or smartphone, with a graphic display, such as a color display, mounted inside the operator's cab of the apparatus 111. The cockpit computer 115 may implement some or all of the operations and functions further described herein for the mobile computer apparatus 104. Networks 109 broadly represent any combination of one or more data communication networks, including local area networks, wide area networks, internal networks, or the internet, using any wired or wireless links, including terrestrial or satellite links. Networks can be implemented through any medium or mechanism that allows data exchange between the various elements in Figure 1. The various elements in Figure 1 can also have direct communication links (wired or wireless). The sensors 112, the controller 114, the external data server computer 108, and other system elements each comprise an interface compatible with Networks 109 and are programmed or configured to use standardized protocols for communication across networks, such as TCP / IP, Bluetooth, CAN protocol, and higher-layer protocols such as HTTP, TLS, and the like. The agricultural intelligence computer system 130 is programmed or configured to receive field data 106 from the field manager computing device 104, external data 110 from the external data server computer 108, and sensor data from the remote sensor 112. The agricultural intelligence computer system 130 may be further configured to host, use, or run one or more computer programs, other software elements, digitally programmed logic such as FPGAs or ASICs, or any combination thereof to perform the translation and storage of data values, the construction of digital models of one or more crops in one or more fields, the generation of recommendations and notifications, and the generation and sending of application controller command sequences 114, in the manner further described in other sections of this disclosure. In one embodiment, the agricultural intelligence computer system 130 is programmed with or comprises a communication layer 132, presentation layer 134, data management layer 140, hardware / virtualization layer 150, field data repository and model 160, and code instructions 180. “Layer”, in this context, refers to any combination of digital interface electronic circuits, microcontrollers, firmware such as actuators, and / or computer programs or other software elements. The communication layer 132 can be programmed or configured to perform input / output interface functions, including sending requests to the field manager computing device 104, external data server computer 108, and remote sensor 112 for field data, external data, and sensor data, respectively. The communication layer 132 can also be programmed or configured to send the received data to the field data repository and model 160 for storage as field data 106. Code 180 instructions may include a set of programming code instructions which, when executed by one or more computer processors, cause the processors to perform an approach to generate an enhanced field anomaly map using digital images and machine learning models. In one embodiment, Code 180 instructions comprise image calibration instructions 136, image stitching instructions 137, grid generation instructions 138, and image classification instructions 139. Image calibration instructions 136 can be configured to perform image calibration on raw images such as aerial raw images, UAV raw images, ground images, and the like. Image calibration can include enhancing or correcting image color, brightness, saturation, and similar parameters. It can also include image gamma correction and pixel correction of pixels in the image that appear incorrect or inconsistent. Image Pasting Instructions 137 can be configured to paste, or connect, multiple images into one large image. Pasting can include determining the edges of each image within the multiple images, correcting the edges if necessary to perform accurate pasting, and concatenating the images into one coherent, large image. The grid generation instructions 138 can be configured to generate a grid template for an image, such as a pasted image. In one mode, the grid can include a plurality of rectangles arranged in rows and columns to span the entire image. In another mode, the grid can include a plurality of hexagons, or other shapes, that cover the entire image. Image Classification Instructions 139 can be configured to apply one or more image classifiers to an image. An image classifier can be an image, or a thumbnail image that displays a sample of, for example, an anomaly. Examples of anomalies include a bare soil anomaly, a bending anomaly, a weed anomaly, a standing water anomaly, and the like. The presentation layer 134 can be programmed or configured to generate a graphical user interface (GUI) to be displayed on the field manager computing device 104, the cabin computer 115, or other computers that are coupled to the system 130 through the network 109. The GUI may include controls for entering data to be sent to the agricultural intelligence computer system 130, generating requests for models and / or recommendations, and / or displaying recommendations, notifications, models, and other field data. The data management layer 140 can be programmed or configured to manage read and write operations involving the repository 160 and other functional elements of the system, including queries and result sets communicated between the functional elements of the system and the repository. Examples of the data management layer 140 include JDBC, SQL Server interface code, and / or HADOOP interface code, among others. The repository 160 can comprise a database. As used here, the term “database” can refer to either a data body, a relational database management system (RDBMS), or both.As used here, a database can encompass any set of data, including hierarchical databases, relational databases, flat-file databases, object-relationship databases, object-oriented databases, distributed databases, and any other structured set of records or data stored on a computer system. Examples of RDBMS include, but are not limited to, databases from Oracle®, MySQL, IBM® DB2, Microsoft® SQL Server, Sybase®, and PostgreSQL. However, any database that supports the systems and methods described here may be used. When field data 106 is not provided directly to the agricultural intelligence computer system via one or more farm machines or farm machine devices that interact with the agricultural intelligence computer system, the user may be prompted, through one or more user interfaces on the user device (served by the agricultural intelligence computer system), to enter such information. In one example, the user can specify identification data by accessing a map on the user device (served by the agricultural intelligence computer system) and selecting specific CLUs that have been graphically displayed on the map.In an alternative mode, user 102 can specify identification data by accessing a map on the user device (served by the agricultural intelligence computer system 130) and drawing field boundaries on the map. Such selection of CLUs or map drawings represent geographic identifiers. In alternative modes, the user can specify identification data by accessing field identification data (provided as form files or in a similar format) from the U.S. Department of Agriculture's Farm and Agriculture Services Agency or another source through the user device and providing such field identification data to the agricultural intelligence computer system. In one exemplary configuration, the 130 agricultural intelligence computer system is programmed to generate and display a graphical user interface comprising a data manager for data entry. After one or more fields have been identified using the methods described above, the data manager can provide one or more graphical user interface widgets which, when selected, can identify changes in the field, soil, crops, tillage, or nutrient practices. The data manager can include a timeline view, a spreadsheet view, and / or one or more editable programs. Figure 5 illustrates an exemplary timeline view for data entry. Using the display shown in Figure 5, a user computer can input a selection from a particular field and a particular date for event addition. Events displayed at the top of the timeline might include ncoonn / i ζπζ / β / υιλι Nitrogen, Planting, Practices, and Soil. To add a nitrogen application event, a user computer can provide input to select the nitrogen tab. The user computer can then select a location on the timeline for a particular field to indicate a nitrogen application in the selected field. In response to receiving a selection of a location on the timeline for a particular field, the data administrator can display a data entry overlay, allowing the user computer to enter data pertaining to nitrogen applications, planting procedures, soil application, tillage procedures, irrigation practices, or other information related to the particular field.For example, if a user computer selects a portion of the timeline and indicates a nitrogen application, then the data entry overlay can include fields to enter an amount of nitrogen applied, an application date, a type of fertilizer used, and any other information related to the nitrogen application. In one mode, the data manager provides an interface for creating one or more programs. “Program,” in this context, refers to a set of data pertaining to nitrogen applications, planting procedures, soil applications, tillage procedures, irrigation practices, or other information that may be related to one or more fields and that can be stored in a digital data repository for reuse as an adjustment operation. After a program has been created, it can be conceptually applied to one or more fields, and references to the program can be stored in the digital repository in association with data that identifies the fields.Therefore, instead of manually entering identical data related to the same nitrogen applications for multiple different fields, a user computer can create a schedule that specifies a particular nitrogen application and then apply that schedule to multiple different fields. For example, in the timeline shown in Figure 5, the two top timelines have the “Applied in Spring” schedule selected, which includes an application of 68.03 kilograms (150 pounds) of nitrogen in early April. The data manager can provide an interface for editing a schedule. In one mode, when a schedule is edited, each field selected by that schedule is edited. For example, in Figure 5, if the “Applied in Spring” schedule is edited to reduce the nitrogen application to 58.96 kilograms (130 pounds) of N / ac, the two upper fields can be updated with a reduced nitrogen application based on the edited program. In one mode, in response to receiving edits to a field that has a selected program, the data manager removes the field's correspondence to the program ncoonn / i ζπζ / ε / υιλι 1. For example, if a nitrogen application is added to the upper field in Figure 5, the interface may update to indicate that the "Applied in Spring" program is no longer being applied to the upper field. Although the nitrogen application may remain in early April, updates to the "Applied in Spring" program would not alter the nitrogen application in April. Figure 6 illustrates an example of a spreadsheet view for data entry. Using the display shown in Figure 6, a user can create and edit information for one or more fields. The data administrator can include spreadsheets for entering information regarding Nitrogen, Planting, Practices, and Soil, as shown in Figure 6. To edit a particular entry, a user computer can select the entry in the spreadsheet and update the values. For example, Figure 6 illustrates an update in progress to a target yield value for the second field. Additionally, a user computer can select one or more fields to apply one or more programs. In response to a program selection for a particular field, the data administrator can automatically populate the entries for that field based on the selected program.As with the timeline view, the data administrator can update the entries for each field associated with a particular program in response to a program update. Additionally, the data administrator can remove the selected program's correspondence to the field in response to an edit to one of the field's entries. In one modality, field and model data are stored in a field and model data repository. Model data comprises data models created for one or more fields. For example, a crop model might include a digitally constructed model of crop development in one or more fields. “Model,” in this context, refers to a digitally stored electronic set of executable instructions and associated data values, which have the ability to receive and respond to a programming call or other digital call, invocation, or request for resolution based on specified input values, to produce one or more stored or calculated output values ​​that can serve as the basis for computer-implemented recommendations, output data displays, or machine control, among other things.Those skilled in the technique find it convenient to express models using mathematical equations, but this form of expression does not reduce the models disclosed here to abstract concepts; rather, each model presented here has a practical application on a computer in the form of stored executable instructions and data that implement the model using the computer. The model may include a model of past events in one or more fields, a model of the current status of one or more fields, and / or a model of predicted events in one or more fields. The field and model data may be stored in in-memory data structures, rows in a database table, flat files or spreadsheets, or other forms of stored digital data. In one embodiment, the digital image processing instructions 135 comprise a set of one or more main memory pages, such as RAM, in the agricultural intelligence computer system 130 into which the executable instructions have been loaded and which, when executed, cause the agricultural intelligence computer system to perform the functions or operations described herein with reference to those modules. The instructions may be in computer-executable code within a CPU's instruction set and may be compiled from source code written in Java, C, C++, Objective-C, or any other human-readable or environment-readable programming language, alone or in combination with scripts in JavaScript, other sequencing languages, and other programming source text.The term “pages” is intended to refer broadly to any region within main memory, and the specific terminology used in a system may vary depending on the memory or processor architecture. Alternatively, the digital image processing instructions 135 may also represent one or more source code files or projects that are digitally stored on a mass storage device such as non-volatile RAM or disk storage, in the agricultural intelligence computer system 130, or a separate repository system. When compiled or interpreted, these files or projects generate executable instructions, which, when executed, cause the agricultural intelligence computer system to perform the functions or operations described herein with reference to those modules.In other words, the figure in the drawing can represent the way in which programmers or software developers organize and arrange the source code for later compilation into an executable, or interpretation into bytecode or the equivalent, for execution by the agricultural intelligence computer system 130. The hardware / virtualization layer 150 comprises one or more central processing units (CPUs), memory controllers, and other devices, components, or elements of a computer system such as volatile or non-volatile memory, non-volatile storage such as a disk, and I / O devices or interfaces as illustrated and described, for example, in relation to Figure 4. Layer 150 may also comprise programmed instructions that are configured to support virtualization, containerization, or other technologies. For the purpose of illustrating a clear example, Figure 1 shows a limited number of instances of certain functional elements. However, in other modalities, there can be any number of such elements. For example, modalities may utilize thousands or millions of different mobile computing devices 104 associated with different users. Furthermore, the system 130 and / or the external data server computer 108 may be implemented using two or more processors, cores, clusters, or instances of physical or virtual machines, configured in a discrete location or placed with other elements in a data center, shared computing facility, or cloud computing facility. 2.2. Overview of the application program In one embodiment, the implementation of the functions described herein using one or more computer programs or other software elements loaded onto and executed using one or more general-purpose computers will result in the general-purpose computers being configured as a machine or computer specially adapted to execute the functions described herein. Furthermore, each of the flowcharts further described herein may serve, alone or in combination with the prose descriptions of processes and functions herein, as algorithms, plans, or directions that may be used to program a computer or logic to implement the functions described.In other words, all the prose text here, and all the figures, together are intended to provide a disclosure of algorithms, plans, or directions that are sufficient to enable an expert to program a computer to perform the functions described herein, in combination with such person's skill and knowledge given the level of skill that is appropriate for inventions and disclosures of this kind. In one mode, user 102 interacts with the agricultural intelligence computer system 130 using the field manager computing device 104 configured with an operating system and one or more application programs or applications; the field manager computing device 104 can also interoperate with the agricultural intelligence computer system independently and automatically under program or logic control, and direct user interaction is not always required. The field manager computing device 104 broadly represents one or more smartphones, PDAs, tablet computing devices, laptop computers, desktop computers, workstations, or any other computing device with the ability to transmit and receive information and perform the functions described herein.The field manager computing device 104 can communicate over a network using a mobile application stored on the field manager computing device 104, and in some modes, the device can be coupled using a cable 113 or connector to the sensor 112 and / or controller 114. A particular user 102 may own, operate, or have and use, in connection with the system 130, more than one field manager computing device 104 at a time. The mobile application can provide client-side functionality over the network to one or more mobile computing devices. In one example, the field management computing device 104 can access the mobile application through a web browser or a local client application (app). The field management computing device 104 can transmit data to and receive data from one or more front-end servers using web-based protocols or formats such as HTTP, XML, and / or JSON, or application-specific protocols. In one example, the data can take the form of user requests and input, as well as field data, on the mobile computing device.In some modalities, the mobile application interacts with location tracking hardware and software on the field manager computing device 104, which determines the location of the field manager computing device 104 using standard tracking techniques such as radio signal multilateration, Global Positioning System (GPS), Wi-Fi positioning system, or other mobile positioning methods. In some cases, location data or other data associated with device 104, users 102, and / or user accounts can be obtained by querying a device operating system or by requesting an application on the device to obtain data from the operating system. In one mode, the field management computing device 104 sends field data 106 to the agricultural intelligence computer system 130 comprising or including, but not limited to, data values ​​representing one or more of: the geographic location of one or more fields, tillage information for one or more fields, crops planted in one or more fields, and soil data extracted from one or more fields. The field management computing device 104 can send field data 106 in response to user input from user 102 specifying the data values ​​for one or more fields. Additionally, the field management computing device 104 can automatically send field data 106 when one or more of the data values ​​become available to the field management computing device 104.For example, the field manager computing device 104 may be communicatively coupled to the remote sensor 112 and / or application controller 114, which includes an irrigation sensor and / or irrigation controller. In response to receiving data indicating that the application controller 114 released water onto one or more fields, the field manager computing device 104 may send field data 106 to the agricultural intelligence computer system 130 indicating that water was released onto one or more fields. The field data 106 identified in this disclosure may be entered and communicated using electronic digital data that is communicated between computing devices using parameterized URLs over HTTP, or another suitable messaging or communication protocol. A commercial example of a mobile application is CLIMATE FIELDVIEW, commercially available from The Climate Corporation, San Francisco, California. The CLIMATE FIELDVIEW application, or other applications, may be modified, extended, or adapted to include features, functions, and programming not disclosed prior to the date of this disclosure. In one embodiment, the mobile application comprises an integrated software platform that enables a farmer to make fact-based decisions for their operation by combining historical data pertaining to the farmer's fields with any other data the farmer wishes to compare. The combinations and comparisons can be run in real time and are based on scientific models that provide potential scenarios to enable the farmer to make better, more informed decisions. Figures 2A and 2B illustrate two views of an exemplary logical organization of instruction sets in main memory when an exemplary mobile application is loaded for execution. In Figures 2A and 2B, each item mentioned represents a region of one or more pages of RAM or other main memory, or one or more blocks of disk storage or other non-volatile storage, and the instructions programmed within those regions. In one embodiment, in Figure 2A, a mobile computer application 200 comprises count-fields-data-consumption-share instructions 202, overview and alert instructions 204, digital map book instructions 206, seed and planting instructions 208, nitrogen instructions 210, weather instructions 212, field health instructions 214, and execution instructions 216. In one embodiment, a mobile computer application 200 comprises account instructions, fields, data consumption, and sharing 202, which are programmed to receive, translate, and assimilate field data from third-party systems via manual upload or APIs. Data types may include field boundaries, yield maps, as-planted maps, soil test results, as-applied maps, and / or management zones, among others. Data formats may include form files, third-party native data formats, and / or Farm Management Information System (FMIS) exports, among others. Data reception may occur via manual upload, email with attachment, external APIs that push data to the mobile application, or instructions that call external system APIs to pull data into the mobile application. In one embodiment, the mobile computer application 200 comprises a data mailbox.In response to receiving a selection from the data mailbox, the 200 mobile computer application can deploy a graphical user interface to manually upload data files and import uploaded files to a data manager. In one mode, the Digital Map Book Instructions 206 comprise field map data layers stored in the device's memory and are programmed with data visualization tools and geospatial field notes. This provides farmers with convenient, first-hand information for reference, monitoring, and visual insights into field performance. In another mode, the Alert and Overview Instructions 204 are programmed to provide a broad operational view of what is important to the farmer and timely recommendations for taking action or focusing on particular problems. This allows the farmer to focus time on what needs attention, saving time and preserving yield throughout the season.In one modality, the 208 seed and planting instructions are programmed to provide tools for seed selection, hybrid placement, and script creation, including variable rate (VR) script creation, based on scientific models and empirical data. This allows farmers to maximize yield or return on investment through optimized seed purchase, placement, and population. In one mode, the script generation instructions 205 are programmed to provide an interface for generating scripts, including variable-rate (VR) fertility scripts. The interface allows farmers to create scripts for field implements, such as nutrient applications, planting, and irrigation. For example, a planting script interface might include tools for identifying a seed type for planting. Upon receiving a seed type selection, the mobile computer application 200 can display one or more separate fields in management zones, such as field map data layers created as part of digital map book instructions 206.In one mode, management zones comprise soil areas along with a panel that identifies each soil area and a soil name, texture, drainage for each area, or other field data. The Mobile Computer Application 200 can also deploy tools for editing or creating, such as graphical tools for drawing management zones, such as soil areas, on a map of one or more fields. Planting procedures can be applied to all management zones, or different planting procedures can be applied to different subsets of management zones. When a script is created, the Mobile Computer Application 200 can make the script available for download in a format readable by an application driver, such as an archived or compressed format.Additionally and / or alternatively, a command sequence can be sent directly to the cockpit computer 115 from the mobile computer application 200 and / or uploaded to one or more data servers and stored for further use. In one mode, Nitrogen 210 instructions are programmed to provide tools for informing nitrogen decisions by visualizing nitrogen availability to crops. This allows farmers to maximize yield or return on investment through optimized nitrogen application throughout the growing season.Examples of programmed functions include displaying images such as SSURGO images to enable zone plotting and / or fertilizer application images generated from subfield soil data, such as data obtained from sensors, at high spatial resolution (as fine as millimeters or smaller depending on the proximity and resolution of the sensor); loading existing farmer-defined zones; providing a plant nutrient availability graph and / or map to enable the tuning of nitrogen applications across multiple zones; issuing command sequences to drive machinery; tools for mass data input and adjustment; and / or maps for data visualization, among others.“Mass data entry,” in this context, can mean the entry of data once and then the application of that same data to multiple fields and / or zones that have been defined in the system; exemplary data may include nitrogen application data that is the same for many fields and / or zones of the same farmer, but such mass data entry applies to the entry of any type of field data in the mobile computer application 200. For example, nitrogen instructions 210 can be programmed to accept nitrogen application definitions and practice programs and accept user input specifying the application of those programs across multiple fields.“Nitrogen application programs,” in this context, refers to stored data sets that associate: a name, color code, or other identifier; one or more application dates; types of material or product for each date and quantity; application or incorporation method, such as injected or broadcast; and / or application rates or quantities for each date; and the crop or hybrid being applied, among other things. “Nitrogen practice programs,” in this context, refers to stored data sets that associate: a practice name; a previous crop; a tillage system; a primary tillage date; one or more previous tillage systems that were used; and one or more application type indicators, such as fertilizer, that were used.The Nitrogen 210 instructions can also be programmed to generate and cause the display of a nitrogen chart, which indicates projections of plant use of the specified nitrogen and whether a surplus or deficit is predicted; in some modes, different color indicators can signal the magnitude of the surplus or the magnitude of the deficit.In one embodiment, a nitrogen chart comprises a graphic display on a computer screen device comprising a plurality of rows, each row associated with and identifying a field; data specifying what crop is planted in the field, the size of the field, the location of the field, and a graphical representation of the perimeter of the field; in each row, a timeline by month with graphical indicators specifying each nitrogen application and amount at points correlated with the names of the month; and numerical and / or color indicators of surplus or deficit, wherein the color indicates the magnitude. In one mode, the nitrogen chart may include one or more user input features, such as disks or sliders, to dynamically change nitrogen planting and practice schedules so that a user can optimize their nitrogen chart. The user can then use their optimized nitrogen chart and the related planting and practice schedules to implement one or more scripts, including variable-rate (VR) fertility scripts. Nitrogen 210 instructions can also be programmed to generate and trigger the display of a nitrogen map, which indicates projections of plant use of the specified nitrogen and whether a surplus or deficit is predicted; in some modes, different color indicators may signal the magnitude of the surplus or deficit.The nitrogen map can display projections of plant use of the specified nitrogen and whether a surplus or deficit is predicted for different times in the past and future (such as daily, weekly, monthly, or annually) using numerical and / or color-coded indicators of the surplus or deficit, where the color indicates the magnitude. In one mode, the nitrogen map may include one or more user input features, such as dials or sliders, to dynamically change planting schedules and nitrogen practices so that a user can optimize their nitrogen map, such as to achieve a preferred amount of surplus or deficit. The user can then use their optimized nitrogen map and related nitrogen planting and practice schedules to implement one or more scripts, including variable-rate (VR) fertility scripts.In other modalities, instructions similar to the nitrogen 210 instructions could be used for the application of other nutrients (such as phosphorus and potassium), the application of pesticides, and irrigation programs. In one mode, the Weather 212 instructions are programmed to provide recent field-specific weather data and forecast weather information. This allows farmers to save time and have an efficient, integrated deployment for daily operational decisions. In one mode, Field Health Instructions 214 are programmed to provide timely remote sensing imagery that highlights crop variation in season and potential concerns. Examples of programmed functions include cloud checking to identify potential clouds or cloud shadows; determination of nitrogen indices based on field imagery; graphical visualization of scan layers, including, for example, those related to field health, and viewing and / or sharing scan notes; and / or downloading satellite imagery from multiple sources and prioritizing images for the farmer, among others. In one mode, the 216 performance instructions are programmed to provide reports, analyses, and insight tools using on-farm data for evaluation, insights, and decision-making. This allows the farmer to seek improved results for the following year through fact-based conclusions regarding why the return on investment was at previous levels and an understanding of yield-limiting factors. The 216 performance instructions can be programmed to communicate via the 109 networks to back-end analytical programs running on the farm intelligence computer system 130 and / or an external data server computer 108 and configured to analyze metrics such as yield, yield differential, hybrid, population, SSURGO zone, soil test properties, or elevation, among others.Scheduled reports and analyses may include yield variability analysis, treatment effect estimation, comparative yield analysis and other metrics against other farmers based on anonymized data collected from many farmers, or data for seeds and planting, among others. Applications with instructions configured in this way can be deployed for different computing device platforms while maintaining the same overall user interface appearance. For example, the mobile application can be programmed to run on tablets, smartphones, or server computers accessed using browsers on client computers. Furthermore, the mobile application, as configured for tablet or smartphone computers, can provide a full application experience or a cockpit application experience that is suitable for the deployment and processing capabilities of the cockpit computer.For example, referring now to Figure 2B, in one mode, a cockpit computer application 220 may comprise cockpit-mapping instructions 222, remote-view instructions 224, data collection and transfer instructions 226, machine alert instructions 228, script transfer instructions 230, and cockpit-scanning instructions 232. The codebase for the instructions in Figure 2B may be the same as for Figure 2A, and executables implementing the code may be programmed to detect the type of platform on which they are running and expose, through a graphical user interface, only those functions appropriate for a cockpit or full-platform environment. This approach enables the system to recognize the different user experience appropriate for an in-cockpit environment and the different cockpit technology environment.The cab-map instructions 222 can be programmed to provide map views of fields, farms, or regions that are useful in directing machine operation. The remote view instructions 224 can be programmed to power on, manage, and provide real-time or near-real-time machine activity views to other computing devices connected to the system 130 via wireless networks, wired connectors or adapters, and similar means. The data collection and transfer instructions 226 can be programmed to power on, manage, and provide transfer of data collected by sensors and controllers to the system 130 via wireless networks, wired connectors or adapters, and similar means. The machine alert instructions 228 can be programmed to detect problems with machine or tool operations associated with the cab and generate operator alerts.The 230 script transfer instructions can be configured to transfer instruction scripts that are configured to direct machine operations or data collection. The 232 cabin scan instructions can be programmed to deploy location-based alerts and information received from System 130 based on the location of the Field Manager Computing Device 104, Agricultural Apparatus 111, or sensors 112 in the field, and to assimilate, manage, and provide the transfer of location-based scan observations to System 130 based on the location of the Agricultural Apparatus 111 or sensors 112 in the field. 2.3. Data assimilation to the computer system In one embodiment, the external data server computer 108 stores external data 110, including soil data representing soil composition for one or more fields and climate data representing temperature and precipitation in one or more fields. The climate data may include past and present climate data as well as forecasts for future climate data. In one embodiment, the external data server computer 108 comprises a plurality of servers hosted by different entities. For example, a first server may contain soil composition data while a second server may contain climate data. Additionally, soil composition data may be stored on multiple servers.For example, one server may store data representing the percentage of sand, sediment, and clay in the soil, while a second server may store data representing the percentage of organic matter (OM) in the soil. In one configuration, remote sensor 112 comprises one or more sensors programmed or configured to produce one or more observations. Remote sensor 112 can be aerial sensors, such as satellites, vehicle sensors, planting equipment sensors, tillage sensors, fertilizer or insecticide application sensors, mower sensors, and any other implement capable of receiving data from one or more fields. In one configuration, application controller 114 is programmed or configured to receive instructions from the agricultural intelligence computer system 130. Application controller 114 can also be programmed or configured to control an operating parameter of a vehicle or agricultural implement.For example, an application controller can be programmed or configured to control an operating parameter of a vehicle, such as a tractor, planting equipment, tillage equipment, fertilizer or insecticide equipment, mower equipment, or other farm implements such as a water valve. Other configurations can use any combination of sensors and controllers; the following are simply selected examples. System 130 can obtain or assimilate data under the control of user 102, on a bulk basis from a large number of farmers who have contributed data to a shared database system. This method of obtaining data can be termed “Manual Data Assimilation” insofar as one or more user-controlled computer operations are requested or triggered to obtain data for use by System 130. As an example, the commercially available CLIMATE FIELDVIEW application from The Climate Corporation, San Francisco, California, can be operated to export data to System 130 for storage in repository 160. For example, the seed monitoring system can control components of the planter apparatus and obtain planting data, including signals from seed sensors, through a signaling device comprising a CAN bus and point-to-point connections for recording and / or diagnostics. Seed monitoring systems can be programmed or configured to display seed separation, population, and other information to the user via the in-cab computer. 115 or other devices within system 130. Examples are disclosed in U.S. Patent No. 8,738,243 and U.S. Patent Publication 20150094916, and this disclosure assumes knowledge of those other patent disclosures. Similarly, performance monitoring systems may contain performance sensors for the mower apparatus that send performance measurement data to the cab computer 115 or other devices within the system 130. Performance monitoring systems may use one or more remote sensors 112 to obtain grain moisture measurements on a combine harvester or other mower and transmit these measurements to the user via the cab computer 115 or other devices within the system 130. In one embodiment, examples of sensors 112 that can be used with any moving vehicle or apparatus of the type described elsewhere in this document include kinematic sensors and position sensors. Kinematic sensors may comprise any of the velocity sensors, such as wheel speed sensors or radar, accelerometers, or gyroscopes. Position sensors may comprise GPS receivers or transceivers, or Wi-Fi-based mapping or positioning applications that are programmed to determine location based on nearby Wi-Fi hotspots, among other things. In one embodiment, examples of 112 sensors that can be used with tractors or other moving vehicles include engine speed sensors, fuel consumption sensors, area counters or distance counters that interface with GPS or radar signals, PTO (power take-off) speed sensors, tractor hydraulic sensors configured to detect hydraulic parameters such as pressure or flow, and / or hydraulic pump speed, wheel speed sensors, or wheel slip sensors. In one embodiment, examples of 114 controllers that can be used with tractors include hydraulic steering controllers, pressure and / or flow controllers; hydraulic pump speed controllers; speed controllers or regulators; hitch position controllers; or wheel position controllers that provide automatic steering. In one modality, examples of 112 sensors that can be used with seed planting equipment such as planters, drills, or air seeders include seed sensors, which may be optical, electromagnetic, or impact sensors; downforce sensors such as load pins, load cells, pressure sensors; soil property sensors such as reflectivity sensors, moisture sensors, electrical conductivity sensors, optical residue sensors, or temperature sensors; component operating criteria sensors such as planting depth sensors, downforce cylinder pressure sensors, seed disc speed sensors, seed drive motor encoders, seed conveyor system speed sensors,or vacuum level sensors; or pesticide application sensors such as optical or other electromagnetic sensors, or impact sensors. In one embodiment, examples of 114 controllers that may be used with such seed planting equipment include: toolbar folding controllers, such as controllers for valves associated with hydraulic cylinders; downforce controllers, such as controllers for valves associated with pneumatic cylinders, air bags, or hydraulic cylinders, and programmed to apply a downforce to individual row units or an entire planter frame; planting depth controllers, such as linear actuators; metering controllers, such as electric seed meter drive motors, hydraulic seed meter drive motors, or row control clutches; hybrid selection controllers,such as seed meter drive motors, or other actuators programmed to selectively permit or prevent seed or an air-seed mixture from being delivered to or from seed meters or central bulk hoppers; metering controllers, such as electric seed meter drive motors, or hydraulic seed meter drive motors; seed conveyor system controllers, such as controllers for a belt-delivery seed conveyor motor; marker controllers, such as a controller for a pneumatic or hydraulic actuator; or pesticide application rate controllers, such as metering drive controllers, orifice size or position controllers. In one embodiment, examples of sensors 112 that can be used with tillage equipment include position sensors for tools such as shanks or discs; tool position sensors for such tools that are configured to detect depth, equipment angle, or side clearance; downforce sensors; or pull-force sensors. In one embodiment, examples of controllers 114 that can be used with tillage equipment include downforce controllers or tool position controllers, such as controllers configured to control tool depth, equipment angle, or side clearance. In one embodiment, examples of sensors 112 that may be used in connection with the apparatus for applying fertilizer, insecticide, fungicide and the like, such as in-plant starter fertilizer systems, subsoil fertilizer applicators, or fertilizer sprayers, include: fluid system criterion sensors, such as flow sensors or pressure sensors; sensors indicating which spray head valves or fluid line valves are open; sensors associated with tanks, such as fill level sensors; supply line sensors at the system level or in sections, or row-specific supply line sensors; or kinematic sensors such as accelerometers placed on spray linkage.In one modality, examples of 114 controllers that can be used with such apparatus include pump speed controllers; pump controllers that are programmed to control pressure, flow, direction, PWM and the like; or position actuators, such as for linkage height, subsurface depth, or linkage position. In one embodiment, examples of 112 sensors that can be used with harvesters include performance monitors, such as impact plate extensometers or position sensors, capacitive flow sensors, load sensors, weight sensors or torque sensors associated with elevators or augers, or optical or other electromagnetic grain height sensors; grain moisture sensors such as capacitive sensors; grain loss sensors, including impact, optical or capacitive sensors; headland operating criteria sensors such as headland height, headland type, deck plate clearance, feeder speed, and reel speed sensors; separator operating criteria sensors, such as concave clearance, rotor speed, shoe clearance, or screen clearance sensors; auger sensors for position, operation or speed; or motor speed sensors.In one modality, examples of 114 controllers that can be used with mowers include headland operating criteria controllers for elements such as headland height, headland type, deck plate spacing, feeder speed, reel speed; separator operating criteria controllers for features such as concave clearance, rotor speed, shoe clearance, or screen clearance; or controllers for auger position, operation, or speed. In one embodiment, examples of sensors 112 that can be used with grain carts include weight sensors or sensors for position, operation, or auger speed. In another embodiment, examples of controllers 114 that can be used with grain carts include controllers for position, operation, or auger speed. In one embodiment, examples of sensors 112 and controllers 114 may be installed on an unmanned aerial vehicle (UAV) or “drones.” Such sensors may include cameras with detectors effective for any range of the electromagnetic spectrum, including visible light, infrared, ultraviolet, near-infrared (NIR), and the like; accelerometers; altimeters; temperature sensors; humidity sensors; pitot tube sensors or other wind speed or air velocity sensors; battery life sensors; or radar emitters and reflected radar energy detection apparatus; other electromagnetic radiation emitters and reflected electromagnetic radiation detection apparatus.Such controllers may include motor or guidance control apparatus, control surface controllers, camera controllers, or controllers programmed to ignite, operate, obtain data from, manage, and configure any of the foregoing sensors. Examples are disclosed in U.S. Patent Application No. 14 / 831,165, and this disclosure assumes knowledge of that other patent disclosure. In one embodiment, sensors 112 and controllers 114 can be attached to the soil measuring and sampling apparatus that is configured or programmed to sample soil and perform soil chemistry tests, soil moisture tests, and other soil-related tests. For example, the apparatus disclosed in U.S. Patent No. 8,767,194 and U.S. Patent No. 8,712,148 can be used, and this disclosure assumes knowledge of those patent disclosures. In one embodiment, sensors 112 and controllers 114 may comprise weather devices for monitoring weather conditions in fields. For example, the apparatus disclosed in U.S. Provisional Application No. 62 / 154,207 filed April 29, 2015, U.S. Provisional Application No. 62 / 175,160 filed June 12, 2015, U.S. Provisional Application No. 62 / 198,060 filed July 28, 2015, and U.S. Provisional Application No. 62 / 220,852 filed September 18, 2015, may be used, and this disclosure assumes knowledge of those patent disclosures. 2.4. General perspective of the process - agronomic model training In one mode, the agricultural intelligence computer system 130 is programmed or configured to create an agronomic model. In this context, an agronomic model is a data structure in the memory of the agricultural intelligence computer system 130 that comprises field data 106, such as identification data and harvest data for one or more fields. The agronomic model may also include calculated agronomic properties that describe either the conditions that can affect the growth of one or more crops in a field, or properties of one or more fields, or both. Additionally, an agronomic model may include recommendations based on agronomic factors, such as crop recommendations, irrigation recommendations, planting recommendations, fertilizer recommendations, fungicide recommendations, pesticide recommendations, harvest recommendations, and other crop management recommendations.Agronomic factors can also be used to estimate one or more crop-related outcomes, such as agronomic yield. The agronomic yield of a crop is an estimate of the amount of crop produced, or in some examples, the income or profit obtained from the crop produced. In one mode, the 130 agricultural intelligence computer system can use a pre-configured agronomic model to calculate agronomic properties related to currently received crop and location information for one or more fields. The pre-configured agronomic model is based on previously processed field data, including but not limited to identification data, harvest data, fertilizer data, and weather data. The pre-configured agronomic model may have been cross-validated to ensure model accuracy. Cross-validation may include comparison with on-the-ground verification that compares predicted results with actual results in a field, such as comparing an estimated precipitation with a rain gauge or sensor that provides weather data at the same or a nearby location, or an estimated nitrogen content with a soil sample measurement. Figure 3 illustrates a programmed process through which the agricultural intelligence computer system generates one or more preconfigured agronomic models using field data provided by one or more data sources. Figure 3 can serve as an algorithm or instructions for programming the functional elements of the agricultural intelligence computer system 130 to execute the operations described below. In block 305, the agricultural intelligence computer system 130 is configured or programmed to implement the processing of agronomic data from field data received from one or more data sources. The field data received from one or more data sources may be pre-processed for the purpose of removing noise, distortion effects, and confounding factors within the agronomic data, including measured outliers that could affect the received field data values ​​in various ways.Modalities of agronomic data preprocessing may include, but are not limited to, removing data values ​​commonly associated with outliers, specific measured data points known to unnecessarily bias other data values, data smoothing, aggregation, or sampling techniques used to remove or reduce additive or multiplicative noise effects, and other data filtering or derivation techniques used to provide clear distinctions between positive and negative data inputs. In block 310, the agricultural intelligence computer system 130 is configured or programmed to perform data subset selection using preprocessed field data to identify datasets useful for generating the initial agronomic model. The agricultural intelligence computer system 130 can implement data subset selection techniques including, but not limited to, a genetic algorithm method, an all-model subset method, a sequence search method, a step-level regression method, a particle pest optimization method, and an ant colony optimization method. For example, a genetic algorithm selection technique uses an adaptive heuristic search algorithm, based on evolutionary principles of natural selection and genetics, to determine and evaluate datasets within the preprocessed agronomic data.In block 315, the agricultural intelligence computer system 130 is configured or programmed to implement the evolution of field datasets. In one mode, a specific field dataset is evaluated by creating an agronomic model and using quality thresholds specific to that model. Agronomic models can be compared and / or validated using one or more comparison techniques, such as, but not limited to, root mean square error with cross-validation leaving one out (RMSECV), mean absolute error, and mean percentage error. For example, RMSECV can cross-validate agronomic models by comparing predicted agronomic property values ​​created by the model against collected and analyzed historical agronomic property values.In one modality, agronomic dataset evaluation logic is used as a feedback loop, where agronomic datasets that do not meet configured quality thresholds are used during the future dataset subset selection steps (block 310). In block 320, the agricultural intelligence computer system 130 is configured or programmed to implement the creation of the agronomic model based on cross-validated agronomic datasets. In one mode, the agronomic model criterion can implement multi-variate regression techniques to create pre-configured agronomic data models. In block 325, the agricultural intelligence computer system 130 is configured or programmed to store the pre-configured agronomic data models for future field data evaluation. 2.5. Implementation of the hardware overview example According to one approach, the techniques described here are implemented by one or more special-purpose computing devices. These special-purpose computing devices may be hardwired to execute the techniques, or they may include digital electronic devices such as one or more application-specific integrated circuits (ASICs) or field-programmable gate arrays (FPGAs) that are persistently programmed to execute the techniques, or they may include one or more general-purpose hardware processors programmed to execute the techniques according to program instructions in firmware, memory, other storage, or a combination thereof. Such special-purpose computing devices may also combine conventional hardwired logic, ASICs, or FPGAs with conventional programming to achieve the techniques.Special purpose computing devices can be desktop computer systems, laptop computer systems, handheld devices, networking devices, or any other device that incorporates program logic and / or is wired to implement the techniques. For example, Figure 4 is a block diagram illustrating a computer system 400 on which an embodiment of the invention can be implemented. The computer system 400 includes a bus 402 or other communication mechanism for transmitting information, and a hardware processor 404 coupled with the bus 402 for processing information. The hardware processor 404 can be, for example, a general-purpose microprocessor. The 400 computer system also includes a main memory 406, such as random access memory (RAM) or another dynamic storage device, coupled to the bus 402 to store information and instructions to be executed by the processor 404. The main memory 406 can also be used to store temporary variables or other intermediate information during the execution of instructions to be executed by the processor 404. Such instructions, when stored on a non-transient storage medium accessible to the processor 404, make the 400 computer system a special-purpose machine that is customized to execute the operations specified in the instructions. The 400 computer system also includes a read-only memory (ROM) 408 or other static storage device coupled to the bus 402 to store static information and instructions for the processor 404. A storage device 410, such as a magnetic disk, optical disk, or solid-state drive, is provided and coupled to the bus 402 to store information and instructions. The computer system 400 can be coupled via bus 402 to a display 412, such as a cathode ray tube (CRT), to display information to a computer user. An input device 414, including alphanumeric and other keys, is coupled to bus 402 to communicate information and command selections to the processor 404. Another type of user input device is the cursor control 416, such as a mouse, a ball, or cursor direction keys, to communicate direction information and command selections to the processor 404 and to control the movement of the cursor on the display 412. This input device typically has two degrees of freedom on two axes, a first axis (e.g., x) and a second axis (e.g., y), which allows the device to specify positions in a plane. The computer system 400 can implement the techniques described herein using custom-hardwired logic, one or more ASICs or FPGAs, firmware, and / or program logic, which, in combination with the computer system, cause or program the computer system 400 to be a special-purpose machine. In one embodiment, the techniques herein are executed by the computer system 400 in response to the processor 404 executing one or more sequences of one or more instructions contained in main memory 406. Such instructions can be read into main memory 406 from another storage medium, such as storage device 410. The execution of the instruction sequences contained in main memory 406 causes the processor 404 to execute the steps of the process described herein.In alternative modalities, hardwired circuits can be used instead of or in combination with software instructions. The term “storage medium” as used herein refers to any non-transient medium that stores data and / or instructions that cause a machine to operate in a specific manner. Such storage media may comprise non-volatile and / or volatile media. Non-volatile media include, for example, optical disks, magnetic disks, or solid-state drives, such as the storage device 410. Volatile media include dynamic memory, such as main memory 406. Common forms of storage media include, for example, a floppy disk, a hard disk, a solid-state drive, magnetic tape, or any other magnetic data storage medium, a CD-ROM, any other optical data storage medium, any physical medium with patterns of holes, RAM, a PROM, an EPROM, a FLASH-EPROM, NVRAM, any other memory chip or cartridge. The storage medium is distinct from, but can be used in conjunction with, transmission media. Transmission media are involved in the transfer of information between storage media. For example, transmission media include coaxial cables, copper wire, and fiber optics, including the wires comprising the 402 bus. Transmission media can also take the form of acoustic or light waves, such as those generated during radio wave communications and infrared data transmission. ncoonn / i ζπζ / β / υιλι Various media can be involved in delivering one or more sequences of instructions to the 404 processor for execution. For example, instructions may initially be carried on a magnetic disk or solid-state drive of a remote computer. The remote computer may load the instructions into its dynamic memory and send them over a telephone line using a modem. A local modem for a 400 computer system may receive the data on the telephone line and use an infrared transmitter to convert the data into an infrared signal. An infrared detector may receive the data carried as an infrared signal, and appropriate circuitry may place the data on the 402 bus. The 402 bus carries the data to the 406 main memory, from which the 404 processor retrieves and executes the instructions.Instructions received by main memory 406 can optionally be stored in storage device 410 either before or after execution by processor 404. The 400 computer system also includes a 418 communication interface coupled to the 402 bus. The 418 communication interface provides a two-way data communication connection to a 420 network link that is connected to a 422 local area network. For example, the 418 communication interface might be an Integrated Services Digital Network (ISDN) card, cable modem, satellite modem, or a modem to provide a data communication connection to a corresponding type of telephone line. As another example, the 418 communication interface might be a local area network (LAN) card to provide a data communication connection to a compatible LAN. Wireless links can also be implemented. In any such implementation, the 418 communication interface sends and receives electrical, electromagnetic, or optical signals that carry digital data streams representing various types of information. Network link 420 typically provides data communication across one or more networks to other data devices. For example, network link 420 might provide a connection across the local network 422 to a host computer 424 or to data equipment operated by an Internet Service Provider (ISP) 426. The ISP 426, in turn, provides data communication services across the worldwide packet data communication network now commonly referred to as the “Internet” 428. The local network 422 and the Internet 428 use electrical, electromagnetic, or optical signals that carry digital data streams. The signals across the various networks and the signals on network link 420 and across the communication interface 418, which carries digital data to and from the computer system 400, are exemplary forms of transmission media. The 400 computer system can send messages and receive data, including program code, through networks, the 420 network link, and the ncoonn / i ζπζ / ε / υιλι communication interface 418. In the internet example, a server 430 could transmit a requested code for an application program over the Internet 428, the !SP 426, the local network 422, and the communication interface 418. The received code can be executed by the processor 404 as it is received, and / or stored in storage device 410, or other non-volatile storage for later execution. 3. Digital Image Processing Approach This paper describes methods that provide computer-implemented techniques for digital image processing of agricultural field images, including strain detection, anomaly detection, and prediction and correction of field data. These methods are most useful in later stages of the growing season and subsequent crop development, when crop and weed cover can be observed in aerial images. However, these techniques can also be used at any stage of a growing season. In some approaches, a large number of digital images of agricultural fields are acquired to train machine learning models, which can then be used to classify specific images captured from agricultural fields during the growing season. Image quality control and preprocessing can be implemented to generate accurate soil data for use in training a machine learning model. Models based on merging classifier output and vegetation index data, such as NDVI or CCL data, can be used. As a result, a digital graph or visual map of anomalies in the fields can be generated. The images can be correlated with actual field data after collection for further validation or calibration.The approaches here can be integrated into a larger data processing workflow for cloud storage or results publication and can be integrated with other models such as performance prediction models. The proposed methodology is based on a combination of selected image generation hardware and innovative image processing algorithms implemented in computer programs. With this method, spatial and spectral image features are identified using high-resolution images received from airborne platforms, unmanned aerial vehicles (UAVs), and ground-mounted vehicle cameras. The images may include color images (such as RGB images) and / or multispectral images (such as near-infrared images). High resolution, in this context, may mean coverage of less than 1 cm per pixel. Machine learning models can be run on feature data to differentiate intact maize plots from previously described field anomalies, and yield classification output is generated. Classified images and image patches can be used to generate geographically rectified maps of intact and non-intact areas of agricultural fields. The maps can be color-coded using colors that correspond to different types of field anomalies. A map can display one or more anomalies. A high-resolution anomaly map generated using these approaches can benefit placement trials, side-by-side field trials, crop protection trials, equipment problem detection, identification of wind damage or puddle formation, yield data adjustment, and quantification of environmental impact at the sub-field level. The presented approach allows for the detection and calculation of all anomalies and their percentages within each mosaic / graph / grid generated for a field. 3.1. Digital image processing for aerial images Figure 7 illustrates an example of digital image processing to generate a field anomaly map using machine learning models. In Figure 7, field data such as boundary data, planting data, yield data, and so on can be stored in a 702 database. The database can be organized as a relational database or another type of database. The database can be hosted as a distributed database system or a standalone server database system. Field metadata describing field boundaries, and all other information shown in Figure 7 using a 704 element, can be provided to aerial vehicles such as helicopters, agricultural aircraft, control centers managing helicopter and aircraft routes, and so on. This information can be used to navigate drones or other unmanned aerial vehicles and direct them to collect aerial images of the field. Upon receiving time-limit information, a 706 aerial vehicle can begin capturing multiple images as it traverses the field. Aerial images can also be obtained from satellites or other aerial vehicles. Images captured 708 are referred to here as aerial / UAV images. These images can be provided as input to a machine learning model 712. The machine learning model 712 can perform image calibration, image processing, and image classification. The machine learning model 712 can also generate a map showing field anomalies based on the images 708. The output of model 712 may include one or more anomaly maps 714. Anomaly maps may include color-coded regions, where each color code expands to a different classification. Examples of classifications include areas that are covered by corn stalks, areas that appear as bare soil, areas that are covered by brush, areas that are covered by roads, and so on. In one mode, a Shapefile 716 map is generated based on map 714. Shapefile 716 can include geographic coordinates of boundaries for one or more areas identified as having anomalies. Map 716 can be provided to ground systems. 3.2. Digital image processing for soil images In one mode, soil systems can use map 716 to control ground cameras to collect soil images of a plot, or plots, identified using the boundaries included in map 716. The soil systems can then use the collected soil images to generate an enhanced plot anomaly map for the plot. The enhanced plot anomaly map displays anomaly details at a higher level of detail than the 714 anomaly map generated from the aerial images described earlier. A ground system may include several cameras, such as camera 718, several sensors, such as sensor / cameras 720, different image capture devices 722, amplifiers 726, and other software / hardware processing tools configured to capture the images. The software and hardware tools are referred to in Figure 7 as an element 724. Ground sensors and cameras can be used to collect soil images according to the Shapefile boundaries provided in Shapefile 716. The Shapefile can include geographic coordinates that specify the boundaries of a farm plot. Therefore, a combine harvester that has, for example, a camera installed on one of its arms, can move across a farm plot according to the coordinates provided in the Shapefile, and as the harvester moves across the plot, it can collect soil images of the plot. In one mode, soil images are calibrated, pre-processed, and stitched together to form a resulting 728 image. The 728 image can include a representation of the plot that is covered by corn, some weeds, some bare soil, and the like. A set of images 728 can be transferred as input to a model 730. The model 730 can be implemented as a machine learning model and can perform various functions, such as processing all the images provided by the ground systems and calibrating the images. This can include, for example, boundary adjustment, color adjustment, hue adjustment, gamma component adjustment, and so on. The model 730 can also process these images. This can include stitching and other processing that will be described later. The resulting images can be processed to determine the classification of individual regions within the image. This classification allows for the identification of which areas or portions of the field are covered by corn stalks, which areas are covered by weeds, which areas are covered by soil, bare soil, and so on. Output images, as described later, can include an anomaly set map, and each map can display an individual anomaly, such as bare soil, weeds, and so forth. The maps can also be supplied to a 702 database. 4. Exemplary processing of aerial and UAV images An aerial survey is a method for collecting geomatic data using data collection instruments installed on airplanes, helicopters, UAVs, balloons, and other mobile devices. Examples of geomatic data can include aerial imagery, lidar data, images representing various visible and invisible bands of the electromagnetic spectrum, geophysical data, and similar data. An aerial survey can also refer to the analysis of charts or maps of geographic regions. Aerial surveys typically provide data at a higher resolution than, for example, data provided by satellites. The proposed approach consists of an image acquisition stage and a machine learning stage. In the image acquisition stage, for a ground-based image generation platform, a custom computer system comprising a Raspberry Pi processor, camera, and GPS receiver automatically acquires georeferenced RGB digital images during harvesting or other agricultural field operations. In this context, georeferencing means that each digital image, at the time of capture and storage, is stored in association with geolocation metadata, i.e., a Shapefile. The metadata can include latitude and longitude values ​​obtained from a GPS receiver mounted on the device with the camera and processor.Retrieving geo-location data and storing location data with images allows for reconstructing a complete image of a field later and / or generating digital maps based on the execution of the machine learning stage using the collected images. Additionally or alternatively, for a UAV-based imaging platform, a high-resolution color camera (e.g., the Sony RX1R-II) or a multispectral camera is integrated with a commercial drone platform (e.g., the MD4-1000 Microdrones or DJI M600), which can be programmed to automatically inspect a predefined area and collect high-resolution, multispectral color images. The use of mobile discovery image generation platforms of this type allows for the collection of data and images as one moves through the field. Subfield zones can be identified in completed images for high-precision image generation and detection. For example, subfield zone metadata can be added to images at the time of image collection, if a zone map is available in computer memory at the time of image capture. GPS data obtained from a GPS receiver can be used to correlate zone maps with the current location of a UAV or harvester capturing images. Furthermore, the hardware arrangements proposed here can reduce the cost and development time required to scale image generation capabilities. Figure 8 illustrates an example of aerial and UAV image processing to generate a field anomaly map using machine learning models. In Figure 8, one or more aerial and UAV images (802) are fed to a calibration unit (804). This calibration may include color correction, hue correction, resolution correction, gamma correction, and so on. The calibrated and pre-processed images are fed to a stitcher (806) that stitches the calibrated images together at a field level. A field-level map refers to an image that covers a typical US agricultural field, which is, for example, 40 to 100 acres in size. The field-level map is generally defined by its boundaries. In stark contrast, a parcel-level map refers to a small rectangular area within the field, which might be, for example, 2 acres wide and 6.09 meters (20 feet) long. Typically, several hundred raw or processed images are stitched onto a field-level image, which is usually a large orthomosaic. An example of an orthomosaic is an 808 image. In one mode, image 808 is fed to a grid generator that divides the orthomosaic image into a grid of small spatial grids. Each grid might be, for example, 64 by 64 pixels to cover an area of ​​3.04 meters by 3.04 meters (10 feet by 10 feet) square. These details are provided for illustrative purposes only and should not be considered a limitation in any way. Actual spatial grids may be larger or smaller. This depends on the implementation. The field grid 810, or a set of small spatial grids, is then converted into a small-tile image grid 812. In a next step, small tiles 812 are provided to a sorting unit, and the images may include, for example, an image 814, an image 816, and so on. A classification and post-processing unit 818 can use a machine learning model, such as the model 712 described in Figure 7. The output of model 818 may include one or more anomaly maps. The maps may include maps 822. The content of maps 822 may be displayed according to a legend 824, which describes different colors assigned to different classified regions. One region may correspond, for example, to corn stalks, while other images may show weeds, or bare soil, and so on. In one mode, based on the images, model 818, called a shapefile, is generated. An example of a shapefile is a Shapefile 820. The Shapefile 820 can include, for example, geographic coordinates for different regions, for different grids or small tiles, which include the classified feature. 5. Exemplary soil image processing These techniques can also be used to generate elapsed-time images of a field by repeatedly capturing images of the field at different times using a camera-computer apparatus mounted in a fixed location within the field, such as a pole. In one embodiment, an elongated pole is placed in the ground in a field, and a solar cell array and computer chassis are attached to the pole. The chassis is fixed in an elevated location so that a camera on the apparatus has a clear view of the field from a high vantage point. The solar cell array is coupled to the computer chassis to serve as a power supply. The computer chassis comprises a Pi camera, a Raspberry Pi 2B processor, a solar panel controller, and an LTE modem.The processor can be programmed to instruct the camera to capture an image every hour and to power the LTE modem to upload the images to cloud data storage periodically. Figure 9 illustrates an example of soil image processing to generate a field anomaly map using machine learning models. Figure 9 is a detailed example of soil processing. It assumes that a Shapefile 916 is provided to ground vehicles, such as mowers, harvesters, and tractors, which are equipped with cameras configured to collect images such as field soil images. The images can be more detailed than aerial / UAV images. In one mode, ground images can be captured by cameras, such as a 718 camera that can be mounted on a combine harvester, tractor, planter, and similar equipment. Other cameras may include the 720, 722, and 724 cameras, which can be mounted on poles, fences, and similar structures. In one modality, ground processing includes amplifying ground images executed, for example, by an amplifier 728, or any other processing element 726. Ground images can be fed into a 730 model, which is configured to collect, calibrate, and process the images. Processing may include running image classification to determine if the images show any anomalies in the field. Output 732 of model 730 can include one or more anomaly maps 734. The maps, as described earlier, can be organized as a map set, and each map can indicate a separate or individual anomaly. For example, one map might show vegetation, another might show bare soil, and so on. The maps can be stored in database 702. Alternatively, or additionally, the maps can be used or transmitted to ground vehicles and agricultural machinery to control the vehicles and machines in performing various farming operations. For example, if one of the anomaly maps indicates areas of a farm field that are covered with bare soil but should be planted with seeds, then the map can be sent to a seeder to instruct it to plant the seeds in those areas. 6. Exemplary implementation of soil image processing In one embodiment, a computer-camera apparatus is attached to the arm of a combine harvester or mower. The apparatus may comprise a Raspberry Pi processor, a Pi camera, a U-Blox GPS board, and a Wi-Fi adapter. In this embodiment, the processor is programmed to instruct the camera to capture an image when the current geographic location of the mobile combine harvester or mower, as determined by reading the location data from the GPS board, coincides with a prescription for image capture. Prescriptions or programs for image capture may specify image capture when the combine harvester is passing over particular points in space, or using a specified separation distance as the combine harvester moves across the field, or according to other schemes. In one scenario, using images captured from a mower as described above, approximately 300 individual images were manually labeled; approximately 230 images were labeled to indicate normal, undamaged plots, good crop positions, and visible walks, and approximately 70 images were labeled to indicate gaps and bending. A CNN transfer learning model was developed using Inception v.3 in TensorFlow and Domino. This model achieved 91% prediction accuracy with N=35. Examples of normal and abnormal plots are shown in the figures and / or specification sheets. In one mode, crop images were captured using a combine-mounted camera based on GPS or distance trigger signals transmitted to the camera from the Raspberry Pi processor. Thumbnail images were produced and wirelessly transmitted as GeoTIFF images to a gateway computer mounted on the combine. The gateway transmitted the image data to cloud storage using wireless transmission and was also programmed to retrieve Shapefiles from cloud storage and upload them via a microUSB connection to the Raspberry Pi processor. A Trimble GPS receiver provided geolocation data and generated a geolocation log that was uploaded to cloud storage.The geolocation data, in combination with the image data, underwent image stitching to combine images captured at adjacent positions in the field as the combine harvester moved, and post-processing to remove artifacts, adjust vertical orientation, and so on. The resulting processed images were then used for model development, training, validation, and classification as described here. In another method, images were captured using a radio-controlled, ground-traveling exploration robot or other unmanned ground vehicle equipped with a Ublox GPS receiver and a Raspberry Pi 2B camera. This device was able to traverse a field and capture images within the field primarily for identifying the location of diseased plants or crop damage. In another approach, an underleaf disease imaging system was used, consisting of a ZED stereo camera mounted on a short pole in a field and connected to an NVidia TX1 computer with a waterproof enclosure. The ZED is a color stereo camera capable of capturing 2K UHD images at 30 frames per second. The TX1 computer is battery-powered and included a second camera. A CHC RTX GPS receiver was mounted separately on another pole and connected to the computer. This device was capable of capturing more than 8,000 images of Goss's wilt, gray leaf spots, and common mold. In yet another application, a 16-channel Velodyne VLP-16 LiDAR device was mounted on a mobile harvester and was able to generate bending images in cornfields. Bend values ​​greatly affect crop yield, yet human visual classifications are labor-intensive to obtain and slow. Digital imaging can increase the efficiency and measurement of all test plots during experiments or treatment comparisons when equipped on combine harvesters. In one configuration, this device was programmed to image four rows of corn on the left side of the harvester. A Garmin GPS was communicatively coupled to the LiDAR, allowing wireless transmission of the LiDAR image data to cloud-based servers. 6.1 Exemplary Edge Computing Implementation Edge computing often refers to data processing that occurs close to the data sources. In imaging applications, edge computing devices are typically deployed on image collection platforms located near the cameras and sensors, rather than on a centralized computing server in the cloud. Edge computing generally helps an imaging system reduce unnecessary data traffic between the system and the central database or cloud, and provides real-time image processing capabilities. An AI accelerator, or neural network accelerator, is an application-specific integrated circuit (ASIO) designed to support artificial neural networks, machine vision systems, and machine learning systems. Examples of vendors that have developed their own AI accelerators include the Intel-based Nervana Neural Network Processor (NNP), the Google-based Tensor Processing Unit (TPU), and the Nvidia-based Graphics Processing Unit (GPU). The Edge TPU, for example, is a solution developed by Google that combines the advantages of edge computing and AI accelerators. In other words, the Edge TPU is a small, low-power solution that can be deployed on an image-generating device powered by, for example, a battery or generator.The Edge TPU can help the image generation system improve AI computing capabilities and provide a platform for running a machine learning / AI model in pseudo-real time. ncoonn / i ζπζ / ε / υιλι In one modality, an approach to mapping field anomalies using digital images and machine learning models is implemented using edge computing technologies. Examples of edge computing technologies have already been described. One such technology is Edge TPU technology. However, the approach presented here is not limited to Edge TPU implementation. In fact, other approaches can also be implemented. Figure 10 illustrates an exemplary ground image processing method for generating a field anomaly map using machine learning models and an Edge TPU. Figure 10 illustrates a particular implementation of the process shown in Figure 9. A Shapefile 1012 is provided to ground systems, and the ground systems use Shapefile 1012 to determine field boundaries and control ground cameras to collect ground images of the field. Subsequently, the collected ground images are processed using, for example, an Edge TPU hardware unit 1006 that is communicating with a communication gateway 1024. In one mode, an on-ground system 1016 can use a Raspberry Pi 2 processor 1018 and a GPS trigger that is generated based on the Shapefile 1012. The trigger is sent to cameras installed on the ground vehicles to instruct the cameras to take raw images 1014 of some areas of the field. Images 1014 captured by cameras mounted on machinery or vehicles on the ground can be sent as JPEG images to the Edge TPU 1006 for processing. The Edge TPU 1006 can apply one or more classifiers to the images to perform image classification. The images can be sent via Ethernet or provided via USB 2.0 devices, for example, as miniature TIFF images 1022, to the gateway 1024. The images 1022 can also be sent (1020) from the gateway 1024 to the processor 1018 for further processing. Gateway 1024 can be implemented as a server or a computer processor and can send classified images as thumbnails 1026, for example, in TIFF format to a cloud system 1004. The TIFF images stored in the cloud system 1004 can also be stored in the database 1010. 7. Exemplary machine learning approach In the machine learning stage, in one modality, deep machine learning (transfer) models programmed based on the pre-trained ImageNet convolutional neural network model (Inception v3) are programmed to classify digital images into multiple categories. The first category, in one modality, is intact rows of crops, such as corn or similar. The second category is non-intact rows of corn that occur due to bending, weeds, and / or bare soil. The model output is used to generate a map of the areas in field images where each image is classified as intact corn, bending, weeds, and bare soil.Although crop bending or damage, weeds, and bare soil are identified here for the purpose of providing a clear example, other modalities can operate to classify images for other anomalies, such as burning, animal damage, heat damage, and so on, based on one or more training datasets that have been selected and used to train the CNN to address those anomalies. Figure 11 illustrates an exemplary machine learning approach for classifying images to generate a field anomaly map using machine learning models. In Figure 11, input images, such as Image 1102, are provided to a machine learning model 1104 that, among other things, performs image classification. The classification may involve the use of a variety of classifiers. In one mode, the classifiers can include a plurality of several image samples showing known anomalies. Examples of anomalies might include inter-plot damage, weeds, standing water, and the like. For each type of anomaly, one or more classifiers can be provided. In Figure 11, the classifiers show inter-plot damage and include an Inter-plot Image #1, an Inter-plot Image #2, and the like. The images for the same anomaly can include different images of the same anomaly, and each image might show, for example, a different view of the anomaly, a different subtype of the anomaly, a different color scheme used to display the anomaly, and so on. The classification process can use images that determine whether an in-ground image illustrates anomalies such as weeds, trees, and the like. To perform image classification, the process can use several classifier images, such as inter-plot damage image #1, inter-plot damage image #2, weed image #1, weed image #2, weed image #2, and so on. All the images can be different. Therefore, when input image 1102 is subjected to the classification process 1104, the classifiers are applied to the grid tiles of input image 1102 to determine if image 1102 matches any of the classifiers. The decision is referred to as output 1106, and may include detailed information regarding whether image 1102 matches any of the classifiers, and if so, whether the matched classifier is inter-plot damage image #1, inter-plot damage image #2, weed image #1, weed image #2, weed image #2, and similar ones. ncoonn / i ζηζ / Β / γ 8. Exemplary classifiers In one modality, an inventory of 5,000 to 6,000 images was obtained and classified to train a machine learning model. Classification labels can include CORN, INTER-ROW DAMAGE, ROAD, SOIL, SOYBEAN, TREES, WATER, WEEDS, SHADE, CONSTRUCTION, but other labels could also be used in other modalities based on the content of the image inventory. In one modality, digital images captured from aerial equipment are programmatically fed into a calibration stage where, for example, image artifacts can be removed, pixel sizes can be normalized, and other preprocessing is performed. The images can then be divided into Level 1 grids consisting, for example, of 640 x 640 pixel tiles. Each tile can be a multi-pixel array of a portion of a source image. In one modality, a plurality of times are then selected for training or validation. Level 2 grid fixing can be applied using 64 x 64 pixel tiles. Other modalities can use grid fixing with different pixel dimensions. In one modality, level 2 grid tiles are manually labeled for soil, weeds, inter-row spacing, and other features. These labeled tiles are then used to train a convolutional neural network for classification or are otherwise used for model development and implementation. Subsequently, the trained model can be used to perform classification on other raw digital image files obtained from aerial or other equipment, alone or in combination with vegetation index data such as NDVL data. When a combination is used, the vegetation index data for a particular field is merged or blended with classification output for digital images of the same field and processed programmatically to generate a field anomaly map. Figure 12 illustrates an exemplary image classification for generating a field anomaly map using machine learning models. In this example, an input image 1202 includes a grid of tiles, each representing either corn, soil, weeds, and so on. Image 1202 is processed by applying a set of classifiers 1204 to the image to determine an output image 1206. Image 1206 may include tile classification, and each tile may have an associated classifier identifier indicating whether the tile corresponds to a particular crop or anomaly. Thus, image 1206 may be classified, for example, as corn, a road, soil, weeds, water, trees, or similar features. The different types of anomalies are shown as element 1204. 9. Exemplary image classification Figure 13 illustrates an example of image classification using a machine learning approach to generate a field anomaly map using machine learning models. In the example shown, different input images (1300) are processed in a calibration and post-processing unit, and then the pre-processed and calibrated images are subjected, in step 1320, to a classification process using a machine learning model. For example, images 1302 to 1310 can be provided to a calibration and pre-processing system 1320, and once the images are calibrated and pre-processed, the images are classified using the approach described in previous figures. The machine learning model can generate output 1350, which includes the classified image. In Figure 13, the classified digital images include a weed map 1352, a bare soil map 1354, a bending map 1356, and an inter-plot damage map 1358. Other anomaly maps can also be generated. The different types of anomalies depend on specific field characteristics. 10. Example neural network configuration Figure 14 illustrates an example of a neural network configuration for generating a field anomaly map using machine learning models. In the example shown, pseudo-machine code 1402 defines the organization of layers, input variables, blocks, and so on, of the model. The example provided is used for illustrative purposes only, and the actual content of the neural network configuration depends on the specific implementation and field characteristics. In the example shown, code 1402 is organized in such a way that a header 1404 includes a description of the layer, the layer type, the output format, and parameter numbers. For example, one of the layers might be an input layer 1406 that includes different input parameters such as 64 by 64 and three, and the number of parameters here is zero. Another element of the neural network configuration may include a 1408 block, and another element may include a 1410 block. Depending on the implementation, the neural network configuration may be different for different models and different implementations. Typically, the network configuration includes a brief description, such as a brief description 1420 showing the total parameter count. The configuration may also include a count of trainable parameters 1422, and counts of non-trainable parameters 1424. 11. Example flowchart for aerial and UAV image processing Figure 15 illustrates an exemplary flowchart for processing aerial and UAV imagery to generate a field anomaly map using machine learning models. The steps described in Figure 15 can be executed by a distributed computing system deployed in the cloud, on a server, or any other processing system configured to collect, process, and classify images. In step 1502, a processor receives raw aerial / UAV imagery for a field. The imagery can be provided by satellites, helicopters, drones, or any other aerial vehicles configured to collect imagery. In step 1504, the processor calibrates, adjusts, and / or pre-processes the images. As described earlier, this can include adjusting the colors in the images, adjusting color saturation, adjusting resolution, adjusting image formats, performing gamma calibration, and any other type of processing necessary to improve the quality of the raw images. In step 1506, the calibrated, adjusted, and pre-processed images are stitched together to create a field-level map. The stitching process typically involves performing the stitching operation on hundreds of images to generate a large orthomosaic image. This image can be substantial in size, as it may cover a large area of ​​the ground. In step 1508, based on the field-level map, the processor generates a grid map. Grid generation typically involves dividing the large orthomosaic image into a grid of smaller spatial grids, which, for example, might be 64 by 64 pixels, covering 3.04 by 3.04 meter (10 by 10 foot) regions of the field. These numbers may vary and may depend on the implementation. In step 1510, the grid is divided into a plurality of small tiles, each of which, as mentioned earlier, can cover, for example, an area of ​​3.04 by 3.04 meters (10 by 10 feet). In step 1512, using a machine learning model, each of the small grid tiles is classified to determine whether the tile represents an area of ​​the field covered with some anomalies, such as weeds, water, bare soil, and so on. The classification process can be performed using the classifiers described earlier. In step 1514, each of the classified images is post-processed, which may include determining the probability of the classification being correct and creating one or more maps showing the classified tiles. For example, as shown in previous figures, the classified images can be used to generate a map showing the location of weeds in the field. The classified images can also be used to generate another map, and that map can show areas that are only covered with bare soil. Yet another map can be generated to show only the areas that are covered by trees. In step 1516, based on the output generated by the machine learning model, a processor generates a Shapefile. The Shapefile includes geographic coordinates (latitude and longitude values) to reference the classified tiles or classified regions in the field. For example, if a weed map is determined based on the classified images, then that map illustrates the areas covered by weeds. Based on that map, a Shapefile can be generated. The Shapefile can provide or include geographic coordinates that define a boundary or boundaries of the weed-covered areas. 12. Example flowchart for soil image processing Figure 16 illustrates an exemplary flowchart for processing ground images to generate a field anomaly map using machine learning models. The steps described in Figure 16 are typically executed by an on-ground system and may utilize advanced hardware technology, such as an Edge TPU. The on-ground processing system can be implemented as a distributed system, such as a cloud system, a virtual system, or a set of autonomous servers. In step 1601, a processor receives a Shapefile that includes geographic coordinates referencing different areas within a field. As described in Figure 15, the Shapefile might include, for example, the geographic coordinates of regions covered with brush, or the Shapefile might include the geographic coordinates of regions covered with trees, and so on. In an alternative approach, the Shapefile can include the boundaries of all anomalies, regardless of their type. For example, the Shapefile can include coordinates of enclosed regions, and one of these regions might be covered by vegetation, another by bare soil, and so on. In step 1602, the processor receives raw soil images for a field. The images can be collected from areas defined by geographic coordinates. As described earlier, the Shapefile can be sent to in-ground vehicles, such as mowers, harvesters, tractors, and the like. Alternatively, or additionally, the Shapefile can be sent to in-ground controllers and / or cameras attached to physical poles placed in the field. The cameras can be triggered or instructed to capture images of different regions. The instructions can provide the geographic coordinates of the particular regions, and the geographic coordinates can be provided in the Shapefile. Raw soil images, for example, can be collected by a tractor as it moves across the field, following the boundaries provided in a Shapefile. In step 1604, the processor calibrates, adjusts, and pre-processes the raw images. This can include color calibration, color and tint adjustment, saturation, gamma correction, image format conversion, and so on. In step 1606, the calibrated, adjusted, and pre-processed images are stitched together to create a large parcel-level map. A parcel-level map refers to a small rectangular area within a field. Small areas might be, for example, 2 parcel-crop widths and 6.09 meters (20 feet) long to cover, for example, 0.002 acres. In contrast, a field-level map refers to an image covering a typical large agricultural field that is, for example, 40 to 100 acres in size. In step 1608, based on the parcel-level map, the processor generates a grid map. Because the grid is generated at a parcel level, it may not cover smaller areas than the grid generated for aerial and UAV imagery. For example, for soil image generation processing, the parcel level might be generated from five to eight images, and these are stitched together into an image coverage of, say, 0.002 acres. In contrast, for aerial / UAV image processing, stitching involved combining several hundred images into a large orthomosaic image coverage of, say, hectares. In step 1610, the map is divided into a plurality of small tiles according to the grid. In step 1612, using a machine learning model, each of the small tiles in the grid is classified. The classification process has been described in previous diagrams and may include matching the tile image to a classifier image. There could be a large set of different classifiers. If a match is found within a certain acceptable probability, then the grid tile is classified based on the matching classifier image. In step 1614, each of the classified images is post-processed, for example, to correct or fill in missing information, and / or correct the classification if the probability is too low. This may also include reclassifying the mosaic image or running the classification again. The post-processed classified images can be used to generate a map showing the classified mosaics. Similarly, as in the aerial / UAV image processing in step 1514 of Figure 15, in step 1614 of Figure 16, the images can be used to generate separate maps, each map representing a different anomaly. For example, one map can be created for an anomaly corresponding to brush, another map for an anomaly corresponding to trees, and so on. One difference between the maps generated in step 1614 of Figure 16 and the maps generated in step 1514 of Figure 15 is that the maps generated in step 1614 have a higher level of precision and granularity and cover a smaller area than the maps generated in step 1514. The maps generated in step 1614 are more specific, precise, and accurate than the maps generated based on satellite and aerial imagery in step 1514 of Figure 15. In step 1616, the post-processed classified images or maps are stored in a database. This may include storing the images in global and / or international data repositories that can be shared among different industries. The images can also be shared among laboratories and research institutions. Additionally, the images can be shared among farmers, agricultural producers, and industries responsible for manufacturing seeds, crops, fertilizers, and agricultural machinery. 13. Benefits of some modalities These modalities provide the ability to identify and map specific anomalies in a crop field using high-throughput imaging with common-color and multispectral imaging sensors, as well as timely mapping of yield-losing areas within a field using low-cost, vehicle-mounted sensors. In the approach proposed here, the use of low-cost sensors combined with machine learning models provides high-quality, high-accuracy maps of various anomaly sources that are scalable to a typical commercial field. The methods assume that a convolutional neural network has been trained, using a large set of digital field images as a training set, to identify image features known to represent crop cover, bare soil, crop damage, and weeds. The models can be trained using images showing crops, bare soil, damaged crops, and weeds in varying proportions with manual labeling of the image meaning. ncoonn / i ζπζ / β / υιλι NOVELTY OF THE INVENTION Having described the present invention, it is considered a novelty and, therefore, priority is claimed for the content contained in the following:

Claims

1. A computer-implemented method for generating an enhanced field anomaly map using digital imagery and machine learning models, characterized in that it comprises: obtaining a Shapefile that defines the boundaries of an agricultural parcel; obtaining a plurality of parcel images from one or more image capture devices located within the boundaries of the agricultural parcel; calibrating and pre-processing the plurality of parcel images to create a parcel map of the agricultural parcel at a parcel level; based on the parcel map, generating a parcel grid; and based on the grid and the parcel map, generating a plurality of parcel tiles.Based on the plurality of parcel mosaics, generate, using a first machine learning model and a plurality of first image classifiers corresponding to one or more first anomalies, a set of classified parcel images showing at least one anomaly; based on the set of classified parcel images, generate a parcel anomaly map for the agricultural parcel; transmit the parcel anomaly map to one or more controllers that control one or more agricultural machines to perform agricultural functions on the agricultural parcel.

2. The computer-implemented method according to claim 1, characterized in that the Shapefile is generated by: obtaining a plurality of aerial images of an agricultural field; calibrating and pre-processing the plurality of aerial images to create a field map of the agricultural field at a field level; based on the field map of the agricultural field, generating a field grid; based on the field grid and the field map, generating a plurality of field mosaics; based on the plurality of field mosaics, generating, using a second machine learning model and a plurality of second image classifiers corresponding to one or more second anomalies, a set of classified field images showing at least one anomaly; and based on the set of classified field images, generating a field anomaly map for the agricultural field;Based on the field anomaly map, generate the boundaries for the agricultural plot.

3. The computer-implemented method according to claim 2, characterized in that the agricultural plot is a part of the agricultural field.

4. The computer-implemented method according to claim 2, characterized in that the plot anomaly map has a higher level of detail than the field anomaly map; wherein the plurality of first image classifiers has a higher level of detail than the plurality of second image classifiers; wherein the plurality of first image classifiers includes two or more of: one or more inter-row image classifiers, one or more weed image classifiers, one or more bare soil classifiers, one or more bending classifiers, or one or more standing water classifiers; wherein one or more first anomalies have a higher level of detail than one or more second anomalies.

5. The computer-implemented method according to claim 2, characterized in that the Shapefile is used to control one or more image capture devices configured to capture one or more parcel images from the agricultural parcel defined by the boundaries.

6. The computer-implemented method according to claim 2, characterized in that the plurality of parcel images is captured by one or more image capture devices insofar as one or more image capture devices are controlled based on the content of the Shapefile specifying the boundaries of the agricultural parcel.

7. The computer-implemented method according to claim 2, characterized in that one or more image capture devices are installed on any of: moving agricultural equipment, stationary agricultural equipment, stationary poles, stationary structures, handheld devices, or mobile devices.

8. The computer-implemented method according to claim 1, characterized in that the calibration and pre-processing of the plurality of parcel images to create the agricultural parcel map at the parcel level comprises performing one or more of the following: calibrating the plurality of parcel images, stitching the plurality of parcel images at the parcel level to create the graphic map, or correcting one or more colors displayed in the plurality of parcel images.

9. The computer-implemented method according to claim 1, characterized in that the calibration and pre-processing of the plurality of parcel images to create the agricultural parcel map at the parcel level are performed by an edge tensor processing unit (Edge TPU).

10. The computer-implemented method according to claim 1, characterized in that the parcel anomaly map for the agricultural parcel comprises one or more specific anomaly maps, each specific anomaly map showing a specific anomaly identified for the agricultural parcel.

11. One or more non-transient storage media that store instructions which, when executed by one or more computing devices, cause one or more computing devices to perform the following: obtain a Shapefile that defines the boundaries of an agricultural parcel; obtain a plurality of parcel images from one or more image capture devices located within the boundaries of the agricultural parcel; calibrate and pre-process the plurality of parcel images to create a parcel map of the agricultural parcel at a parcel level; based on the parcel map, generate a parcel grid; based on the grid and the map, generate a plurality of parcel tiles;Based on the plurality of parcel mosaics, generate, using a first machine learning model and a plurality of first image classifiers corresponding to one or more first anomalies, a set of classified parcel images showing at least one anomaly; based on the set of classified parcel images, generate a parcel anomaly map for the agricultural parcel; transmit the parcel anomaly map to one or more controllers that control one or more agricultural machines to perform agricultural functions on the agricultural parcel.

12. One or more non-transient storage media according to claim 11, characterized in that they store additional instructions which, when executed by one or more computing devices, cause one or more computing devices to perform the following: obtain a plurality of aerial images of an agricultural field; calibrate and pre-process the plurality of aerial images to create a map of the agricultural field at a field level; based on the map of the agricultural field, generate a field grid; based on the field grid and the field map, generate a plurality of field mosaics; based on the plurality of field mosaics, generate, using a second machine learning model and a plurality of second image classifiers corresponding to one or more second anomalies, a set of classified field images showing at least one anomaly;Based on the set of classified field images, generate a field anomaly map for the agricultural field; based on the field anomaly map, generate the boundaries for the agricultural plot. 13.- One or more non-transitory storage means in accordance with claim 12, characterized in that the agricultural plot is a part of the agricultural field. 14.- One or more non-transient storage means according to claim 12, characterized in that the parcel anomaly map has a higher level of detail than the field anomaly map; wherein the plurality of first image classifiers has a higher level of detail than the plurality of second image classifiers; wherein the plurality of first image classifiers includes two or more of: one or more inter-row image classifiers, one or more weed image classifiers, one or more bare soil classifiers, one or more bending classifiers, or one or more standing water classifiers; wherein one or more first anomalies have a higher level of detail than one or more second anomalies. 15.- One or more non-transient storage media according to claim 12, characterized in that the Shapefile is used to control one or more image capture devices configured to capture one or more images of parcels of the agricultural parcel defined by the boundaries. 16.- One or more non-transient storage media according to claim 12, characterized in that the plurality of parcel images is captured by one or more image capture devices insofar as one or more image capture devices are controlled based on the content of the Shapefile that specifies the boundaries of the agricultural parcel. 17.- One or more non-transient storage media according to claim 12, characterized in that one or more image capture devices are installed on any of: moving agricultural equipment, stationary agricultural equipment, stationary poles, stationary structures, handheld devices or mobile devices 18.- One or more non-transient storage means according to claim 11, characterized in that the calibration and pre-processing of the plurality of parcel images to create the agricultural parcel map at the parcel level comprises performing one or more of: calibrating the plurality of parcel images, pasting the plurality of parcel images at the parcel level to create the graphic map, or correcting one or more colors displayed in the plurality of parcel images. 19.- One or more non-transient storage means according to claim 11, characterized in that the calibration and pre-processing of the plurality of parcel images to create the agricultural parcel map at the parcel level are performed by an edge tensor processing unit (Edge TPU). 20.- One or more non-transient storage means according to claim 11, characterized in that the parcel anomaly map for the agricultural parcel comprises one or more specific anomaly maps, each specific anomaly map showing a specific anomaly identified by the agricultural parcel.

Images