Aerial image based deep learning seed cotton yield estimation method

Abstract

There is disclosed a computer-implemented method for estimating seed cotton yield. The method comprises providing aerial image data based on one or more aerial images of seed cotton breeding plots; extracting, from the provided aerial image data, seed cotton breeding plot data indicative of one or more of the seed cotton breeding plots; inputting the extracted seed cotton breeding plot data into a convolutional neural network, CNN, the CNN configured to assess seed cotton breeding plots that the extracted seed cotton breeding plot data are indicative of; and obtaining, from the CNN, a value for each assessed seed cotton breeding plot, wherein the value is representative for seed cotton yield; wherein the estimated seed cotton yield comprises or is derived from the one or more values.

Description

[0001] BASF Agricultural Solutions US LLC 240266W001

[0002] 1 B19163WO

[0003] AERIAL IMAGE BASED DEEP LEARNING SEED COTTON YIELD ESTI MATION M ETHOD

[0004] TECHNICAL FIELD

[0005] Disclosed are a computer-implemented method for estimating seed cotton yield, a data processing apparatus for estimating seed cotton yield, a computer program product, a computer-readable medium, and a use of the method.

[0006] TECHNICAL BACKGROUN D

[0007] Commercial cotton breeding operations require harvesting many hundreds of plots at the culmination of each growing season. Harvesting and weighing plots in-situ requires heavily customized and expensive harvest equipment staffed with a crew of skilled personnel. Harvesting breeding plots is a time consuming, labor intensive, inherently dangerous, and destructive means to assess per-plot cotton yield. Hence, there is room and need for improvement regarding the assessment of per-plot cotton yield.

[0008] It is therefore an object of the present disclosure to overcome at least part of the drawbacks available in the assessment of per-plot cotton yield.

[0009] SUMMARY OF THE I NVENTION

[0010] I n view of the above, according to a first aspect of the present disclosure, there is provided a computer-implemented method for estimating seed cotton yield. The method comprises providing aerial image data based on one or more aerial images of seed cotton breeding plots. The method further comprises extracting, from the provided aerial image data, seed cotton breeding plot data indicative of one or more of the seed cotton breeding plots. The method further comprises inputting the extracted seed cotton breeding plot data into a convolutional neural network (CNN), the CNN configured to assess seed cotton breeding plots that the extracted seed cotton breeding plot data are indicative of. And the method further comprises obtaining, from the CN N, a value for each assessed seed cotton breeding plot, wherein the value is representative for seed cotton yield. The estimated seed cotton yield comprises or is derived from the one or more values.

[0011] Said in other words, there is provided a computer-implemented method for cotton breeders to estimate seed cotton yield on a per-plot basis across large breeding field trials in a nondestructive and safe manner using for example contemporary aerial image technology, like drone technology, a preprocessing routine, and a deep learning CN N architecture trained for example from aerial image orthomosaic plot-level extracts gathered over multiple seasons in defoliated cotton breeding field plots.

[0012] The providing of the aerial image data may be based on obtaining, like acquiring or receiving for example, the aerial image data and / or the one or more aerial images from an apparatus, a device, an entity, a database, a system, etc. BASF Agricultural Solutions US LLC 240266W001

[0013] 2 B19163WO

[0014] It should be noted that the seed cotton yield or the estimated seed cotton yield is seed cotton yield per (breeding) plot or estimated seed cotton yield per (breeding) plot.

[0015] Moreover, regarding the applicability of the method according to the first aspect, it should be noted that cotton is chemically defoliated prior to harvest, meaning nearly all leaves are removed from the plant, and therefore the seed cotton is generally visible at the one or more aerial images, i.e. the seed cotton yield is apparent in the images. This may not be the case with crops other than cotton. Said in other words, in other crops the economic product (i.e. the yield) might not even be visible in aerial images. The deep learning CNN architecture is trained on images where the yield is apparent.

[0016] It should further be noted that cotton may represent an example for an agricultural product. In the present disclosure, the term “agricultural product” is to be understood broadly and may, for example, comprise products directly obtained by agricultural harvesting and products containing or derived from products directly obtained by agricultural harvesting. Thus, the agricultural product may comprise raw cotton and one or more products containing or derived from said raw cotton.

[0017] Further, aerial images of seed cotton breeding plots may be images captured by an airborne vehicle, wherein these images may be indicative of the seed cotton breeding plots. In particular, these images may show the seed cotton breeding plots. The air-borne vehicle may have captured these images when flying over the seed cotton breeding plots. The airborne vehicle may be an aircraft, an unmanned aerial vehicle (UAV), a drone, a satellite, or the like. The air-borne vehicle may be an autonomous or a non-autonomous vehicle. It shall be noted that a number of the aerial images based on which the aerial image data are provided may not be limited and may, for example, be any number of aerial images as appropriate or required, for example based on a total amount of images to be captured, based on an (maximum) amount of data to be processed, based on an image quality or pixel resolution to be achieved, or based on a predetermined setting.

[0018] According to several examples of the present disclosure, the extracted seed cotton breeding plot data that are indicative of the one or more of the seed cotton breeding plots may be understood to comprise the one or more of the seed cotton breeding plots, for example all seed cotton breeding plots. Moreover, according to several examples of the present disclosure, the extracted seed cotton breeding plot data may be derived or obtained from high resolution RGB images exclusively. Additionally or alternatively, multispectral images or hyperspectral images could also be used. I.e., the extracted seed cotton breeding plot data may comprise RGB image data. Additionally or alternatively, the extracted seed cotton breeding plot data may comprise multispectral image data or hyperspectral image data.

[0019] The expression “derived from” as used above is to be interpreted in a broad way and may mean that the one or more values obtained from the CNN may be calculated or scaled to real world seed cotton yield values. The one or more values obtained from the CNN may fall between a minimum CNN value, for example zero, and a maximum CNN value, for example one. However, it shall be noted that the minimum and maximum CNN values may be different values. For example, merely for increasing understandability, the minimum and maximum CNN values may be zero and ten. The scaling may be based on a given historical range of seed cotton yields in the CNN’s training data. For example, the given historical range may have a lower limit value and an upper limit value. The lower limit value may be BASF Agricultural Solutions US LLC 240266W001

[0020] 3 B19163WO zero pounds per plot for example. The upper limit value may be 35 pounds per plot for example. However, it shall be noted that the lower and upper limit values may be different values. For example, merely for increasing understandability, CNN training data may be used with a lowest seed cotton yield being five pounds per plot and with a highest seed cotton yield being 50 pounds per plot. The scaling may then comprise that the one or more values obtained from the CNN are scaled to the given historical range. For example, the minimum CNN value is scaled to the lower limit value and the maximum CNN value is scaled to the upper limit value. In more detail, a minimum CNN value of zero may be scaled to a lower limit value of zero pounds per plot and a maximum CNN value of 1 may be scaled to an upper limit value of 35 pounds per plot. All the one or more values between 0 and 1 may then be scaled accordingly to a value between 0 pounds per plot and 35 pounds per plot. For example, a value of 0.5 may then be scaled to a value of 17.5 pounds per plot.

[0021] The term “comprising” does not exclude other elements or steps. Furthermore, the terms “comprising”, “including”, “having” and the like may be used interchangeably herein.

[0022] The term “obtaining”, as used herein, may comprise, for example, receiving from another system, apparatus, CNN, or process; receiving via an interaction with a user; loading or retrieving from storage or memory; measuring or capturing using sensors or other data acquisition apparatuses; deriving from an image, for example by use of image processing.

[0023] The term “data”, as used herein, is to be understood broadly in the present case and represents any kind of data. Data may be single numbers / numerical values, a plurality of a numbers / numerical values, a plurality of a numbers / numerical values being arranged within a list, 2 dimensional maps or 3 dimensional maps, but are not limited thereto.

[0024] Moreover, according to a second aspect of the present disclosure, there is provided a data processing apparatus for estimating seed cotton yield. The data processing apparatus comprises one or more processors being configured to carry out the method according to the first aspect.

[0025] The data processing apparatus may be configured as a distributed data processing apparatus. In particular, the method may be executed by different computing devices of the distributed data processing apparatus. In particular, the distributed data processing apparatus may comprise one or more servers and one or more client devices and the method may be executed in part by at least one of the servers and in part by at least one of the client devices.

[0026] Furthermore, according to a third aspect of the present disclosure, there is provided a data processing system for estimating seed cotton yield. The data processing system comprises the data processing apparatus according to the second aspect and / or comprises means to carry out the method according to the first aspect.

[0027] Further, according to a fourth aspect of the present disclosure, there is provided a farm comprising one or more seed cotton breeding plots, i.e. breeding plots for seed cotton. In general, a breeding plot may be understood as a field of plant, for example an agricultural field. The farm comprising the data processing system according to the third aspect and / or the data processing apparatus according to the second aspect. BASF Agricultural Solutions US LLC 240266W001

[0028] 4 B19163WO

[0029] The term “agricultural field” as used herein is to be understood broadly in the present case and presents any area, i.e., surface and subsurface, of a soil to be treated with a fertilizer product. The agricultural field may be any plant or crop cultivation area, such as a farming field, a greenhouse, or the like. A plant may be a crop, a weed, a volunteer plant, a crop from a previous growing season, a beneficial plant or any other plant present on the agricultural field. According to the present disclosure, in particular, the agricultural field may be a field, in particular a breeding plot, for growing seed cotton and the plant may thus be a seed cotton or cotton. The agricultural field may be identified through field data referring to its geographical location or geo-referenced location data. A reference coordinate, a size and / or a shape may be used to further specify the agricultural field.

[0030] Moreover, according to a fifth aspect of the present disclosure, there is provided a computer program product comprising instructions which, when the program is executed by a computing system, cause the computing system to carry out the method according to the first aspect.

[0031] Furthermore, according to a sixth aspect of the present disclosure, there is provided a computer-readable medium comprising instructions which, when the instructions are executed by a computing system, cause the computing system to carry out the method according to the first aspect. The computer-readable medium may have stored thereon the computer program product according to the fifth aspect. The computer-readable medium may be transitory or non-transitory, volatile or non-volatile

[0032] Further, according to a seventh aspect of the present disclosure, there is provided the use of the method according to the first aspect for at least one of the following:

[0033] - determining when to harvest the seed cotton breeding plots,

[0034] - determining a harvesting schedule for one or more of the seed cotton breeding plots, for example for all seed cotton breeding plots,

[0035] - outputting, to a harvester and / or control device for controlling a harvester and / or to a user an indication when to harvest the seed cotton breeding plots on a field and / or the harvesting schedule,

[0036] - outputting selection support data for making a selection decision in breeding and / or trait development, and

[0037] - outputting grouping support data for grouping seed cotton breeding plots in the field in accordance with seed cotton breeding plot fitness. It shall be noted that a point in time where the method according to the first aspect may be deployed occurs after plots have reached maturity and are assumed to be ready for harvest. The term “fitness” is meant to indicate such point in time.

[0038] Each of the above-outlined first to seventh aspects is advantageous in that it is provided nondestructive, low-cost and inherently safe means of gathering per-plot yield values in cotton breeding plots. Each aspect is further advantageous in that it is further provided for cost savings, enhanced efficiency, and improved worker safety in the instances where the solution as disclosed herein is used in place of traditional cotton breeding plot harvesting methodologies. Each aspect further allows for use to estimate seed cotton yield on per-plot basis with an air-borne vehicle, for example a drone, in cotton breeding field trials. Breeders BASF Agricultural Solutions US LLC 240266W001

[0039] 5 B19163WO are thus enabled to use this technology as disclosed herein over traditional methods of harvesting for selected locations or for individual trials at many locations. A primary use case may be in early generation material such as breeding nurseries where a number of plots can range into tens of thousands or early generation yield trials. The ability to gather seed cotton yield information over trials of this size in near real time with an air-borne vehicle allows the breeding team for two advantages, namely, first to offset a large portion of the end of season workload and, second, to identify high performing plots to be sampled for fiber quality parameters.

[0040] According to various examples of the present disclosure, with reference to the first aspect, the extracting may comprise generating one or more images having substantially rectangular shape and depicting two or more seed cotton breeding plots in an orientation aligned with each other and / or aligned with the sides of the rectangle. However, it shall be noted that an amount or number of generated images may not be limited and that any amount or number of images may be generated as appropriate or required. For example, based on an (maximum) amount of data to be processed, or according to a predetermined setting or according to a certain CNN being used. For example, images may be generated to cover all seed cotton breeding plots of a filed. Moreover, it shall be noted that an amount or number of such depicted seed cotton breeding plots may not be limited and that any amount or number of seed cotton breeding plots may be depicted generated as appropriate or required, as long as the generated image may have a substantially rectangular shape. For example, three, four, or five seed cotton breeding plots may be depicted or even ten or more seed cotton breeding plots may be depicted. For example, two seed cotton breeding plots may be preferred.

[0041] It should be noted that the extracted seed cotton breeding plot data may be indicative of and / or may represent the generated one or more images. Hence, the inputting the extracted seed cotton breeding plot data into the CNN may be understood as inputting the generated one or more images into the CNN.

[0042] The depicting is to be understood as a depicting of the two or more seed cotton breeding plots in the generated one or more images. Hence, the sides of the rectangle are to be understood as the sides of the image having a substantially rectangular shape.

[0043] Hence, due to such generating and such depicting, it is facilitated for the CNN to process the inputted data, and a result to be obtained from the CNN based on the processing may be more accurate, i.e. may comprise less data uncertainty. Thus, reliability of an estimated seed cotton yield may be increased.

[0044] According to various examples of the present disclosure, with reference to the first aspect, the extracting may comprise generating, from the extracted seed cotton breeding plot data, first image data indicative of a first image comprising at least part of the one or more seed cotton breeding plots. The extracting may further comprise generating, from the first image data, second image data indicative of a second image. The second image may result from bisecting the first image into a first half and a second half, and from placing the first half and the second half adjacent to one another, so that the second image may be of a substantial rectangular shape, and so that the second image may comprise seed cotton breeding plots running in a same direction and / or being aligned with the sides of the rectangle, in particular, wherein adjacent sides of the substantial rectangular shape may BASF Agricultural Solutions US LLC 240266W001

[0045] 6 B19163WO differ in length by a maximum of 5%, 10%, 15% or 20% of the longer side. The extracted seed cotton breeding plot data may comprise the generated second image data.

[0046] The term “bisecting” may be understood as dividing the first image having the substantially rectangular shape into a first half having a first substantially rectangular shape and a second half having a second substantially rectangular shape. The first substantially rectangular shape and the second substantially rectangular shape may be substantially the same. For example, for each of the first and second substantially rectangular shapes, the longer side(s) of the rectangle may be twice as long as the shorter side(s) of the rectangle. Thus, when placing the first half and the second half adjacent to one another, so that a longer side of the first rectangle is adjacent to a longer side of the second rectangle, the shorter sides of the first and second rectangles form sides equal in length as each of the longer sides of the first and second rectangles, and the shape resulting from such placing may be a square.

[0047] It should be noted that the extracted seed cotton breeding plot data may be indicative of and / or may represent the generated second image or second image data. Hence, the inputting the extracted seed cotton breeding plot data into the CNN may be understood as inputting the generated second image or second image data into the CNN.

[0048] Hence, due to such generating, it is further facilitated for the CNN to process the inputted data, and a result to be obtained from the CNN based on the processing may be even more accurate, i.e. may comprise less data uncertainty. In particular, such step may reduce an extreme distortion that would occur if a naturally rectangular plot was resized to fit standard machine learning inputs. Thus, reliability of an estimated seed cotton yield may be even further increased.

[0049] According to various examples of the present disclosure, with reference to the first aspect, the substantial rectangular shape is a substantial square shape.

[0050] Hence, data processing is even further facilitated for the CNN, and reliability of an estimated seed cotton yield may be even further increased.

[0051] According to various examples of the present disclosure, with reference to the first aspect, the providing the aerial image data may comprise providing aerial image data acquired by or obtained from a single air-borne vehicle flight, like a drone flight for example.

[0052] The single air-borne vehicle flight may be a single flight of an air-borne vehicle over the seed cotton breeding plots.

[0053] Hence, images over a large area may be captured as desired and / or as appropriate, based on controlling the air-borne vehicle as desired and / or as appropriate.

[0054] According to various examples of the present disclosure, with reference to the first aspect, the providing the aerial image data may comprise providing aerial image data acquired by or obtained from an image capturing satellite.

[0055] The image capturing satellite may have flown over the seed cotton breeding plots once or several times.

[0056] Hence, images of even larger areas may be captured and processed. Thus, efficiency may be even further increased. BASF Agricultural Solutions US LLC 240266W001

[0057] 7 B19163WO

[0058] According to various examples of the present disclosure, with reference to the first aspect, the CNN training data may comprise training image data indicative of historical images that comprise historical seed cotton breeding plots associated with the historical seed cotton yields.

[0059] The historical images may be historical aerial images. Thus, the training data may be historical aerial image data.

[0060] Hence, since the training image data are indicative of such historical images, the training image data may be more accurate as simulated data, and may be specifically selected for certain areas, i.e. certain seed cotton breeding plots. Said in other words, for a CNN to process images captured from certain seed cotton breeding plots, the CNN may be trained based on historical images captured from such certain seed cotton breeding plots in the past.

[0061] According to various examples of the present disclosure, with reference to the first aspect, the method may further comprise scaling the obtained one or more values to respective one or more estimated seed cotton yields based on historical seed cotton yields in the CNN training data. For example, seed cotton yields may be estimated for all (breeding) plots of a field.

[0062] The scaling may be any arbitrary scaling, for example a scaling on a scale from 0 to 1, 0 to 10, or 0 to 100.

[0063] Hence, a result obtained from the CNN may be made comparable to a yield to be estimated, and may be made comparable to historical such values.

[0064] According to various examples of the present disclosure, with reference to the first aspect, the CNN may comprise a CNN architecture, and wherein the CNN architecture may comprise a linear activation node as a final node.

[0065] Hence, the CNN may be used to solve a linear regression problem based on imagery rather than a classification problem or object detection problem based on imagery.

[0066] Optional features of the first aspect may form part of any of the second to seventh aspects, mutatis mutandis.

[0067] The indefinite article “a” or “an” does not exclude a plurality. In addition, the articles “a” and “an” as used herein should generally be construed to mean “one or more” unless specified otherwise or clear from the context to be directed to a singular form.

[0068] Unless specified otherwise, or clear from the context, the phrases “one or more of A, B and C”, “at least one of A, B, and C”, and “A, B and / or C” as used herein are intended to mean all possible permutations of one or more of the listed items. That is, the phrase “A and / or B” means (A), (B), or (A and B), while the phrase “A, B, and / or C” means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C).

[0069] The present disclosure may include one or more aspects, examples or features in isolation or combination whether specifically disclosed in that combination or in isolation. Any optional feature or sub-aspect of one of the above aspects applies as appropriate to any of the other aspects. BASF Agricultural Solutions US LLC 240266W001

[0070] 8 B19163WO

[0071] The above-described aspects will become apparent from, and elucidated with, reference to the detailed description provided hereinafter.

[0072] BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In the following, the present disclosure is further described with reference to the enclosed figures:

[0074] Figure 1 illustrates per-plot aerial image data being extracted and associated with relevant breeding metadata using a combination of aerial image orthomosaic, GIS software, and an aligned plot vector layer.

[0075] Figure 2 according to several examples of the present disclosure, illustrates in a first section (indicated with “1”) an extracted plot that is georeferenced and need to be reoriented to remove “no data” values; indicates in a second section (indicated with “2”) a clipping and reorienting step that may be conducted using a Python script and standard computer vision libraries; and illustrates in a third section (indicated with “3”) a step to bisect the image, i.e the extracted plot as shown in the first and second sections, and place the two halves of the extracted plot adjacent to one another.

[0076] Figure 3 according to several examples of the present disclosure, illustrates in a first section (indicated with “1”) that an initial portion of the architecture handles the convolution layers and is repeated four times in the architecture; and illustrates in a second section (indicated with “2”) a final leg of the architecture that may be a fully connected dense layer with single linear activation node that allows for the image inputs to result in single continuous numerical output.

[0077] Figure 4 illustrates results of model deployment at location-year-combinations according to several examples of the present disclosure.

[0078] Figure 5 illustrates a flowchart indicative of a method according to several examples of the present disclosure.

[0079] Figure 6 shows a block diagram schematically illustrating a data processing apparatus according to several examples of the present disclosure.

[0080] Figure 7 illustrates, as an example, at least part of two breeding plots extracted from an aerial imagen, wherein the breeding plots have a comparatively high yield as derivable from the comparatively many and large white areas that indicate a comparatively high amount of cotton.

[0081] Figure 8 illustrates, as an example, at least part of two breeding plots extracted from an aerial imagen, wherein the breeding plots have a comparatively low yield as derivable from the comparatively few and small white areas that indicate a comparatively low amount of cotton. BASF Agricultural Solutions US LLC 240266W001 9 B19163WO

[0082] DETAILED DESCRIPTION OF EMBODIMENTS

[0083] The following embodiments are mere examples for implementing the method, the data processing apparatus or the data processing system for example and shall not be considered limiting.

[0084] Referring now to Figure 1, Figure 1 illustrates per-plot aerial image data, like drone data being extracted and associated with relevant breeding metadata using a combination of aerial image orthomosaic, like drone orthomosaic, geospatial information system (GIS) software, and an aligned plot vector layer.

[0085] In more detail, according to several examples of the present disclosure, aerial image-based image data from one of many sensor types are collected over breeding plots 100 at various bandwidths. It shall be noted that the specific type of sensor is not important. Said in other words, it is not important, which type of sensor or image generation technology is used, as long as it allows to differentiate between cotton and the plants as required for the present example. In view thereof, the sensor types may comprise one or more of the following examples: an RGB camera to obtain RGB image data, a multispectral camera to obtain multispectral image data, a hyperspectral camera to obtain hyperspectral image data.. The raw imagery is stitched to generate an ultra-high resolution, two-dimensional, georeferenced orthomosaic 101 with pixel resolution, for example in terms of ground sample distance (GSD), for example of several millimeters per pixel, for example below 10 mm / pixel. However, it shall be noted that the lower limit of the pixel resolution may for example be defined by the need to separate the breeding plots, e.g. to identify the borders of each plot. The maximal resolution depends on the amount of data that can be processed, or which additional information shall be collected. For example, the pixel resolution is between 0,01mm / pixel to lOOmm / pixel, in one example, the pixel resolution is between 0,lmm / pixel and 20mm / pixel, like below lOmm / pixel and above lmm / pixel. This enables a visualization of an individual breeding plot 102 for example. A single air-borne vehicle flight, like single drone flight used in this method may be collected after all breeding plots 100 have been chemically defoliated as is common practice in cotton cultivation. The defoliant application intensity and the cultivation environment may dictate the precise timing of the air-borne vehicle flight used in the seed cotton yield estimation method.

[0086] In general, breeding experiments consist of many hundreds of breeding plots almost always organized in a linear arrangement, as indicated by the breeding plots 100 in Figure 1. To associate the air-borne vehicle gathered information with the respective breeding plots in a trial the creation of a plot vector layer is required. The plot vector layer may be created with a GIS tool where the precise alignment to the planted breeding plots 100 in the field space is verified. The GIS tool facilitates embedding relevant per-plot meta data for each of the many individual breeding plots to be analyzed into the plot vector layer. Once the plot vector layer is created the process for extracting a portion of the orthomosaic 101, for example an individual breeding plot 102, s conducted using a script written in the Python programming language for example. The result of the plot extraction process may be many individual plot level raster extractions stored in a directory for further analysis.

[0087] Referring now to Figure 2, according to several examples of the present disclosure, Figure 2 illustrates in a first section 210 (indicated with “1”), a second section 220 (indicated with “2”), and a third section 230 (indicated with “3”). In the first section 210, it is illustrated an BASF Agricultural Solutions US LLC 240266WC01

[0088] 10 B19163WO extracted plot 240, in more detail at least part of two individual breeding plots each one may be understood to represent an individual breeding plot 102 as illustrated above with reference to Figure 1, that is georeferenced and that needs to be reoriented to remove “no data” values, wherein such “no data” values may correspond to such surrounding areas 261, 262, 263 and 264 around the plot 240 (i.e. the extracted plot or the plot to be extracted) as illustrated in the first section 210 and the second section 220. Further, Figure 2 indicates in the second section 220 a clipping and reorienting step that may be conducted using a Python script and standard computer vision libraries. Furthermore, Figure 2 illustrates in the third section 230 a step to bisect the image, i.e the extracted plot 240 as shown in the first (and second) section(s), and to place the two halves 270a and 270b of the extracted plot 240 adjacent to one another to form a new image 270, which may have a substantial square shape.

[0089] In more detail, according to several examples of the present disclosure, after completing the plot extraction step the resultant output may be preprocessed into a form suitable to be consumed by the CNN deep learning architecture. The many hundreds to thousands of raw plot extracts from a given field or trial are produced, for example, using a Python script with a standard GIS processing library. The per-plot extracts may be georeferenced to a real- world coordinate reference system where the orientation is as it appears on the Earth’s surface when viewed from above assuming North is toward the top of the extract, as illustrated as an example for the plot 240 in the first section 210 of Figure 2. A multi-step, preprocessing phase may occur so that the relevant portion of the image can be removed so that, schematically, an uncolored / white rectangle 250 as illustrated as an example in the second section 220 of Figure 2 remains. I.e. the relevant portion is removed from the surrounding "no data" portion, the "no data" portion schematically illustrated by the black areas 261, 262, 263 and 264 around the uncolored / white rectangle 250 as illustrated as an example in the second section 220 of Figure 2. Hence, the resulting image conforms to the input requirements of the machine learning model architecture.

[0090] In more detail, a first step in this process may be to isolate the relevant portion of the image (see the first section 210 of Figure 2 and the isolated (and the extracted) plots 240), clip out the relevant portion (see the second section 220 of Figure 2 and the extracted isolated plots 250), and transform the clipped portion to create a new image 270 (see the third section 230 of Figure 2). This may be performed with a Python script and standard computer vision processing library. The plot level extract may be subjected to a binary thresholding process where each pixel in the image is converted to either 0 or 255 depending on whether it is an image pixel (i.e. a pixel representing and / or corresponding to (part of) a certain plot, for example plot 240) or a no data pixel (i.e. a pixel not representing and / or not corresponding to (part of) a certain plot, for example a pixel representing and / or corresponding to a surrounding area of such certain plot, i.e. area 261, 262, 263 and 264). The precise plot boundaries in the plot extract are identified and the plot is clipped, reoriented, and exported as a non-georeferenced standard image file, as indicated in the second section 220 of Figure 2.

[0091] A further step in the preprocessing phase is an approach to dealing with the non-square plot extracts (in terms of image dimensions) that are common to cotton breeding experimental plots. It is possible to design a CNN that accepts non-square inputs but there BASF Agricultural Solutions US LLC 240266W001

[0092] 11 B19163WO are complications that arise when dealing with such inputs, specifically the creation of nonsquare convolutional kernels to ensure uniform reduction from one layer of the network to the next.

[0093] According to several examples of the present disclosure, to avoid these subtle complexities, the provided method takes an approach in bisecting the breeding plot 240 and then placing the two halves 270a and 270b thereof side by side in the processed image 270. The outcome is much closer to a square in terms of dimensions, as indicated in the third section 230 of Figure 2.

[0094] Then, as a further preprocessing modification step, a resizing or resizing step may be performed, common to machine learning, to ensure all inputs precisely conform to the architecture input size.

[0095] As a general example for improving understandability of what is illustrated in Figure 2, reference is made to Figures 7 and 8. Figure 7 illustrates, as an example, at least part of two individual breeding plots extracted from an aerial imagen, wherein the individual breeding plots have a comparatively high seed cotton yield as derivable from the comparatively many and large white areas (specifically the white cotton lint) that indicate a comparatively high amount of seed cotton. In contrast thereto, Figure 8 illustrates, as an example, at least part of two individual breeding plots extracted from an aerial imagen, wherein the individual breeding plots have a comparatively low seed cotton yield as derivable from the comparatively few and small white areas (specifically the white cotton lint) that indicate a comparatively low amount of seed cotton. Hence, based on what is to be learned from Figures 7 and 8 as comparative examples, the plot 240 according to Figure 2 may be understood to have a comparatively high seed cotton yield.

[0096] Referring now to Figure 3, according to several examples of the present disclosure, Figure 3 illustrates in a first section 310 (indicated with “1”) that an initial portion of the architecture handles the convolution layers and is repeated four times in the architecture; and illustrates in a second section 320 (indicated with “2”) a final leg of the architecture that may be a fully connected dense layer with single linear activation node. That allows for the image inputs to result in single continuous numerical output.

[0097] In more detail, according to several examples of the present disclosure regarding the machine learning model architecture or model architecture, a CNN architecture used in this disclosure begins with a convolution layer followed by an activation layer and a batch normalization layer which is followed by a spatial dropout layer and max pooling layer. Such architecture is illustrated in Figure 3. This layer pattern is repeated four times in the architecture followed by a fully connected dense layer with activation, normalization, and dropout layers. The final node is a linear activation node. Using a linear activation function in the final node is not standard practice in CNN architecture because such architectures typically finish with several nodes with activation functions that will classify imagery such as a SoftMax activation function for example. The present disclosure differs in that it uses a CNN to solve a linear regression problem based on imagery rather than a classification problem or object detection problem based on imagery.

[0098] Referring now to Figure 4, according to several examples of the present disclosure, Figure 4 illustrates results of model deployment at location-year combinations. BASF Agricultural Solutions US LLC 240266W001

[0099] 12 B19163WO

[0100] In more detail, according to several examples of the present disclosure, Figure 4 depicts the performance of the model described in this disclosure. What is described is the relationship between seed cotton yield as evaluated by a mechanical harvester with in-situ weigh system and seed cotton yield as predicted by the deep learning model as a processing method presented in this disclosure. In this instance the model was applied to a location- year-combination never presented during any portion of the training and validation process. Each subplot is a specific trial at this location. The performance of the model varies somewhat depending on the stage of material in the breeding pipeline and therefore the trial evaluated.

[0101] For example, in North America cotton cultivation occurs primarily in the southern to southeastern states in a region commonly referred to as the “Cotton Belt”. The Cotton Belt is divided into three macro regions: Texas Plains, Mid-South (the Mississippi Delta), and the Southeast. As indicated above, for deployment of the model, location-year-combination never presented during any portion of the training and validation process may be used. For example, as a mere example for increasing understandability, in case the model was trained with training data for the year 2022 and the location Texas Plains, the model may be deployed for any year and the location Mid-South or Southeast. Additionally or alternatively, the model may be deployed for any year except 2022 and the location Texas Plains.

[0102] As an alternative or additional approach, the model architecture may be trained on training data specific to a region where the model will be deployed. For example, the model may be trained on training data specific to the Texas Plains and the model will be deployed in the Texas Plains. What may result therefrom is a set of model weights for each macro region within the Cotton Belt for example, but a similar model architecture and preprocessing routine across all regions.

[0103] The model output may be a single value for each plot being assessed that falls between 0 and 1 for example. Given the historical range of seed cotton yields in the training data the values between 0 and 1 may be scaledseed cotton yield values which may range from 0 to 25 pounds per plot or more like from 0 to 35 pounds per plot or even more than 35 pounds per plot. The scaler used can be modified to closer approximate the distribution of yield in a single location based on user expectation of yield variability at that location. In 2023, the model was presented with year-location combinations not seen by the model in training. The locations themselves were used in previous years and therefore are represented in the training data but each growing season is different, meaning the same location in a new year provides an opportunity to deploy the model against entirely unseen data. The results of this practical validation are seen in Figure 4. Hence, Figure 4 illustrates results of model deployment at location-year-combinations not seen at any phase of model development or training. Results on early generation material provided the strongest results. For each of the 15 diagrams or subplots shown in Figure 4, the x axis is labeled “PlotWt Raw” and the y axis is labeled “pred_seed_cotton_wt”. “PlotWt Raw” represents raw plot weight data, i.e. seed cotton yield as evaluated by a mechanical harvester with an in-situ weigh system. “pred_seed_cotton_wt” represents a predicted seed cotton weight, i.e. seed cotton yield as predicted by the deep learning model as a processing method presented in this disclosure. In each subplot, a dotted line represents an angle bisector, i.e. a perfect correlation between “PlotWt Raw” and “pred_seed_cotton_wt”. In each subplot, a dashed line BASF Agricultural Solutions US LLC 240266W001

[0104] 13 B19163WO represents a fitted straight line that is fitted to the single points. Each of such points represents a result obtained from the model in comparison to a corresponding result obtained by the mechanical harvester. The value for the parameter r2 as provided in each subplot may in this case, theoretically, have a value between 0 and 1, wherein a value of 0 may indicate that there is no correlation between “PlotWt Raw” and “pred_seed_cotton_wt”, wherein a value of 1 may indicate that there is a perfect correlation between “PlotWt Raw” and “pred_seed_cotton_wt”, i.e. all points were arranged at the angle bisector. Hence, the larger a value for the parameter r2 is the stronger an obtained result is. Further, each of the subplots comprises two headings, wherein one of the two is either “AGA” or “LGA” that represent different stages of material in the breeding pipeline. The other one of the two headers represents a name or an identification of the material.

[0105] For improving understandability of what is outlined above with reference to Figures 3 and 4, reference is made to Figures 7 and 8 that show two examples of breeding plots with different seed cotton yields as already outlined above in more detail. The machine learning model is trained to learn convolutions to activate portions of the imagery most associated with yield, specifically the white cotton lint in the imagery according to Figures 7 and 8 for example. The only output from the model may be the scaled prediction between 0 and 1 from the image regression problem.

[0106] Referring now to Figure 5, Figure 5 illustrates a flowchart indicative of a method according to several examples of the present disclosure. The method is a computer-implemented method for estimating seed cotton yield.

[0107] The method starts in S500.

[0108] In S510, the method comprises providing aerial image data based on one or more aerial images of seed cotton breeding plots. The seed cotton breeding plots may be such seed cotton breeding plots as illustrated above with reference to Figure 1.

[0109] In S520, the method comprises extracting, from the provided aerial image data, seed cotton breeding plot data indicative of one or more of the seed cotton breeding plots. The extracting may comprise at least of such extracting as outlined above with reference to Figure 2. The extracted seed cotton breeding plot data may be indicative of and / or may comprise such extracted seed cotton breeding plot image 240 as illustrated in Figure 2.

[0110] In S530, the method comprises inputting the extracted seed cotton breeding plot data into a convolutional neural network, CNN, the CNN configured to assess seed cotton breeding plots that the extracted seed cotton breeding plot data are indicative of. The CNN may be such CNN as outlined above with reference to Figure 3.

[0111] In S540, the method comprises obtaining, from the CNN, a value for each assessed seed cotton breeding plot, wherein the value is representative for seed cotton yield. The estimated seed cotton yield comprises or is derived from the one or more values. The one or more values and / or the estimated seed cotton yield may be such values or data points as illustrated in the diagrams of Figure 4. In more detail, the expression “derived from” is to be interpreted in a broad way and may mean that the one or more values obtained from the CNN may be calculated or scaled to real world seed cotton yield values. The one or more values obtained from the CNN may fall between a minimum CNN value, for example zero, and a maximum CNN value, for example one. However, it shall be noted that the minimum BASF Agricultural Solutions US LLC 240266W001

[0112] 14 B19163WO and maximum CNN values may be different values. For example, merely for increasing understandability, the minimum and maximum CNN values may be zero and ten. The scaling may be based on a given historical range of seed cotton yields in the CNN’s training data. For example, the given historical range may have a lower limit value and an upper limit value. The lower limit value may be zero pounds per plot for example. The upper limit value may be 35 pounds per plot for example. However, it shall be noted that the lower and upper limit values may be different values. For example, merely for increasing understandability, CNN training data may be used with a lowest seed cotton yield being five pounds per plot and with a highest seed cotton yield being 50 pounds per plot. The scaling may then comprise that the one or more values obtained from the CNN are scaled to the given historical range. For example, the minimum CNN value is scaled to the lower limit value and the maximum CNN value is scaled to the upper limit value. In more detail, a minimum CNN value of zero may be scaled to a lower limit value of zero pounds per plot and a maximum CNN value of 1 may be scaled to an upper limit value of 35 pounds per plot. All the one or more values between 0 and 1 may then be scaled accordingly to a value between 0 pounds per plot and 35 pounds per plot. For example, a value of 0.5 may then be scaled to a value of 17.5 pounds per plot.

[0113] The method ends in S550.

[0114] According to several examples of the present disclosure, the extracting in S520 may comprise generating one or more images having substantially rectangular shape and depicting two or more seed cotton breeding plots in an orientation aligned with each other and / or aligned with the sides of the rectangle.

[0115] According to several examples of the present disclosure, the extracting in S520 may comprise generating, from the extracted seed cotton breeding plot data, first image data indicative of a first image comprising at least part of the one or more seed cotton breeding plots. The first image may be such extracted seed cotton breeding plot 240 as illustrated in Figure 2. The extracting in S520 may further comprise generating, from the first image data or the first image 240, second image data indicative of a second image. As indicated in Figure 2, the second image may result from bisecting the first image 240 into a first half 270a and a second half 270b, and from placing the first half 270a and the second half 270b adjacent to one another, so that the second image 270 is of a substantial rectangular shape, and so that the second image 270 comprises seed cotton breeding plots running in a same direction and / or being aligned with the sides of the rectangle, in particular, wherein adjacent sides of the substantial rectangular shape differ in length by a maximum of 5%, 10%, 15% or 20% of the longer side. The extracted seed cotton breeding plot data may comprise the generated second image data.

[0116] According to several examples of the present disclosure, the substantial rectangular shape may be a substantial square shape, as indicated by the second image 270 in Figure 2 for example.

[0117] According to several examples of the present disclosure, the providing the aerial image data in S510 may comprise providing aerial image data acquired by or obtained from a single airborne vehicle flight, like drone flight for example. BASF Agricultural Solutions US LLC 240266W001

[0118] 15 B19163WO

[0119] According to several examples of the present disclosure, the providing the aerial image data in S510 may comprise providing aerial image data acquired by or obtained from an image capturing satellite.

[0120] According to several examples of the present disclosure, the CNN training data may comprise training image data indicative of historical images that comprise historical seed cotton breeding plots associated with the historical seed cotton yields, as outlined above with reference to Figure 4 for example.

[0121] According to several examples of the present disclosure, the method according to Figure 5 may further comprise training the CNN with the CNN training data, wherein the CNN training data are specific for a region where an acquirement or obtainment of the aerial image data is to be performed, as outlined above with reference to Figure 4 for example.

[0122] According to several examples of the present disclosure, the method according to Figure 5 may further comprise scaling the from S540 obtained one or more values to respective one or more estimated seed cotton yields based on historical seed cotton yields in the CNN training data, as illustrated in Figure 4 for example.

[0123] According to several examples of the present disclosure, the scaling may comprise using the CNN to solve a linear regression problem for the obtained one or more values, as outlined above with reference to Figure 3 for example.

[0124] According to several examples of the present disclosure, the scaling may comprise using the CNN to solve a linear regression problem for the obtained one or more values, as outlined above with reference to Figure 3 for example.

[0125] Referring now to Figure 6, Figure 6 shows a block diagram schematically illustrating a data processing apparatus 600 according to several examples of the present disclosure. In particular, according to several examples of the present disclosure, there is provided a data processing apparatus 600 for estimating seed cotton yield. The data processing apparatus 600 comprises a processor 601 being configured to carry out the method of Figure 5.

[0126] In more detail, according to various examples, a data processing apparatus 600 being configured to carry out the method of Figure 5 may comprise a processing circuitry, a processing function, a processing means, a processing unit or a processor 601, which enables the data processing apparatus 600 to participate in estimating seed cotton yield. The processor 601 may comprise one or more processing portions or functions, wherein the processing portions or functions may be provided as one or more physical or virtual entities. The data processing apparatus 600 may comprise one or more communication interfaces 602. The data processing apparatus 600 may further comprise a memory or memory unit 603 for storing data, programs and / or instructions to be executed by the processor. The memory 603 may be a memory internal to the data processing apparatus 600 or may be a memory external to the data processing apparatus 600, for example at a cloud server. The processor 601 may comprise one or more portions, which enable the data processing apparatus 600 to execute the method of Figure 5 for example. According to several examples of the present disclosure, a providing portion 610 may be configured to perform such providing according to S510 of Figure 5, an extracting portion 620 may be configured to perform such extracting according to S520 of Figure 5, an inputting portion 630 may be BASF Agricultural Solutions US LLC 240266W001

[0127] 16 B19163WO configured to perform such inputting according to S530 of Figure 5, and an obtaining portion 640 may be configured to perform such obtaining according to S540 of Figure 5.

[0128] According to several examples of the present disclosure, the respective portions of the data processing apparatus 600 may also be understood as means for carrying out the certain function.

[0129] It shall be noted that by the term “processor” 601, it may be referred to an arbitrary logic circuitry configured to perform basic operations of a computer or system, and / or, generally, to a device which is configured for performing calculations or logic operations. In particular, the processor, or computer processor may be configured for processing basic instructions that drive the apparatus, computer or system. It may be a semi-conductor based processor, a quantum processor, or any other type of processor configures for processing instructions. As an example, the processor may comprise at least one arithmetic logic unit ("ALU"), at least one floating-point unit ("FPU)", such as a math coprocessor or a numeric coprocessor, a plurality of registers, specifically registers configured for supplying operands to the ALU and storing results of operations, and a memory, such as an LI and L2 cache memory. In particular, the processor may be a multicore processor. Specifically, the processor may be or may comprise a Central Processing Unit ("CPU"). The processor may be a (“GPU”) graphics processing unit, (“TPU”) tensor processing unit, ("CISC") Complex Instruction Set Computing microprocessor, Reduced Instruction Set Computing ("RISC") microprocessor, Very Long Instruction Word ("VLIW") microprocessor, or a processor implementing other instruction sets or processors implementing a combination of instruction sets. The processing means may also be one or more special-purpose processing devices such as an Application-Specific Integrated Circuit ("ASIC"), a Field Programmable Gate Array ("FPGA"), a Complex Programmable Logic Device ("CPLD"), a Digital Signal Processor ("DSP"), a network processor, or the like. The methods, systems and devices described herein may be implemented as software in a DSP, in a micro-controller, or in any other side-processor or as hardware circuit within an ASIC, CPLD, or FPGA. It is to be understood that the term processor may also refer to one or more processing devices, such as a distributed system of processing devices located across multiple computer systems (e.g., cloud computing), and is not limited to a single device unless otherwise specified.

[0130] Further, the memory 603 may refer to a physical system memory, which may be volatile, non-volatile, or a combination thereof. The memory may include non-volatile mass storage such as physical storage media. The memory may be a computer-readable storage media such as RAM, ROM, EEPROM, CD-ROM, or other optical disk storage, magnetic disk storage, or other magnetic storage devices, non-magnetic disk storage such as solid-state disk or any other physical and tangible storage medium which can be used to store desired program code means in the form of computer-executable instructions or data structures and which can be accessed by the computing system. Moreover, the memory may be a computer-readable media that carries computer- executable instructions (also called transmission media). Further, upon reaching various computing system components, program code means in the form of computer-executable instructions or data structures can be transferred automatically from transmission media to storage media (or vice versa). For example, computer-executable instructions or data structures received over a network or data link can be buffered in RAM within a network interface module (e.g., a “NIC”), and then BASF Agricultural Solutions US LLC 240266W001

[0131] 17 B19163WO eventually transferred to computing system RAM and / or to less volatile storage media at a computing system. Thus, it should be understood that storage media can be included in computing components that also (or even primarily) utilize transmission media.

[0132] According to several examples of the present disclosure, there is provided a data processing system for estimating seed cotton yield. The data processing system comprises the data processing apparatus 600 according to Figure 6 and / or comprises means for carrying out the method according to Figure 5.

[0133] According to several examples of the present disclosure, there is provided a farm comprising the data processing apparatus 600 according to Figure 6 and / or the data processing system as outlined above.

[0134] According to several examples of the present disclosure, there is provided a computer- readable medium comprising instructions which, when executed by a computing system, causes the computing system to perform the method according to Figure 5. The computer- readable medium may be transitory or non-transitory, volatile or non-volatile.

[0135] According to several examples of the present disclosure, there is provided a computer program product comprising instructions which, when executed by a computing system, enable or cause the computing system to perform the method according to Figure 5. The computer program product may comprise a computer-readable medium comprising instructions of the computer program product. The computer-readable medium as mentioned above may have stored thereon the computer program product.

[0136] According to several examples of the present disclosure, there is provided a use of the method according to Figure 5, the data processing apparatus 600, the data processing system as outlined above, the farm as outlined above, the computer-readable medium as outlined above and / or the computer program product as outlined above.

[0137] According to several examples of the present disclosure, the use may comprise at least one of:

[0138] - determining when to harvest the seed cotton breeding plots,

[0139] - determining a harvesting schedule for one or more of the seed cotton breeding plots,

[0140] - outputting, to a harvester and / or control device for controlling a harvester and / or to a user an indication when to harvest the seed cotton breeding plots on a field and / or the harvesting schedule,

[0141] - outputting selection support data for making a selection decision in breeding and / or trait development, and

[0142] - outputting grouping support data for grouping seed cotton breeding plots in the field in accordance with seed cotton breeding plot fitness.

[0143] Optional features of the method according to Figure 5 may form part of the data processing apparatus 600, the data processing system, the farm, the computer-readable medium, the computer program product, and the use, mutatis mutandis.

[0144] Any unit, module, circuitry or methodology described herein may be implemented using hardware, software, and / or firmware configured to perform any of the operations described BASF Agricultural Solutions US LLC 240266W001

[0145] 18 B19163WO herein. Hardware may comprise one or more processor cores, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip systems (SOCs), complex programmable logic devices (CPLDs), etc. Software may be embodied as a software package, code, instructions, instruction sets and / or data recorded on at least one transitory or non- transitory computer readable storage medium. Firmware may be embodied as code, instructions or instruction sets and / or data hard-coded in memory devices (e.g., nonvolatile memory devices).

[0146] If implemented in software, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media include computer-readable storage media. Computer-readable storage media can be any available storage media that can be accessed by a computer. By way of example, and not limitation, such computer-readable storage media can comprise FLASH storage media, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Disk and disc, as used herein, include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray disc (BD), where disks usually reproduce data magnetically and discs usually reproduce data optically with lasers. Further, a propagated signal may be included within the scope of computer-readable storage media. Computer-readable media also includes communications media including any medium that facilitates transfer of a computer program from one place to another. A connection, for instance, can be a communications medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio and microwave are included in the definition of communications medium. Combinations of the above should also be included within the scope of computer-readable media.

[0147] The applicant hereby discloses in isolation each individual feature described herein and any combination of two or more such features, to the extent that such features or combinations are capable of being carried out based on the present specification as a whole in the light of the common general knowledge of a person skilled in the art, irrespective of whether such features or combinations of features solve any problems disclosed herein, and without limitation to the scope of the claims. The applicant indicates that aspects of the present invention may consist of any such individual feature or combination of features.

[0148] It has to be noted that embodiments of the invention are described with reference to different categories. In particular, some examples are described with reference to methods whereas others are described with reference to apparatuses. However, a person skilled in the art will gather from the description that, unless otherwise notified, in addition to any combination of features belonging to one category, also any combination between features relating to different category is considered to be disclosed by this application. However, all features can be combined to provide synergetic effects that are more than the simple summation of the features. BASF Agricultural Solutions US LLC 240266W001

[0149] 19 B19163WO

[0150] While the disclosure has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered exemplary and not restrictive. The disclosure is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art, from a study of the drawings, the disclosure, and the appended claims.

[0151] The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used advantageously.

[0152] Any reference signs in the claims should not be construed as limiting the scope.

Claims

BASF Agricultural Solutions US LLC 240266W00120 B19163WOCLAIMS1. A computer-implemented method for estimating seed cotton yield, comprising: providing (S510) aerial image data based on one or more aerial images of seed cotton breeding plots; extracting (S520), from the provided aerial image data, seed cotton breeding plot data indicative of one or more of the seed cotton breeding plots; inputting (S530) the extracted seed cotton breeding plot data into a convolutional neural network, CNN, the CNN configured to assess seed cotton breeding plots that the extracted seed cotton breeding plot data are indicative of; and obtaining (S540), from the CNN, a value for each assessed seed cotton breeding plot, wherein the value is representative for seed cotton yield; wherein the estimated seed cotton yield comprises or is derived from the one or more values.

2. The computer-implemented method according to claim 1, wherein the extracting comprises generating one or more images having substantially rectangular shape and depicting two or more seed cotton breeding plots in an orientation aligned with each other and / or aligned with the sides of the rectangle.

3. The computer-implemented method according to claim 1 or 2, wherein the extracting comprises generating, from the extracted seed cotton breeding plot data, first image data indicative of a first image comprising at least part of the one or more seed cotton breeding plots; and generating, from the first image data, second image data indicative of a second image, wherein the second image results from bisecting the first image into a first half and a second half, and from placing the first half and the second half adjacent to one another, so that the second image is of a substantial rectangular shape, and so that the second image comprises seed cotton breeding plots running in a same direction and / or being aligned with the sides of the rectangle, in particular, wherein adjacent sides of the substantial rectangular shape differ in length by a maximum of 5%, 10%, 15% or 20% of the longer side, and wherein the extracted seed cotton breeding plot data comprise the generated second image data.

4. The computer-implemented method according to claim 3, wherein the substantial rectangular shape is a substantial square shape.BASF Agricultural Solutions US LLC 240266W00121 B19163WO5. The computer-implemented method according to any of claims 1 to 4, wherein the providing the aerial image data comprises providing aerial image data acquired by a single air-borne vehicle flight.

6. The computer-implemented method according to any of claims 1 to 5, wherein the providing the aerial image data comprises providing aerial image data acquired by an image capturing satellite.

7. The computer-implemented method according to any of claims 1 to 6, wherein the CNN training data comprise training image data indicative of historical images that comprise historical seed cotton breeding plots associated with the historical seed cotton yields.

8. The computer-implemented method according to any of claims 1 to 7, further comprising training the CNN with the CNN training data, wherein the CNN training data are specific for a region where an acquirement of the aerial image data is to be performed.

9. The computer-implemented method according to any of claims 1 to 8, further comprising scaling the obtained one or more values to respective one or more estimated seed cotton yields based on historical seed cotton yields in the CNN training data.

10. The computer-implemented method according to claim 9, wherein the scaling comprises using the CNN to solve a linear regression problem for the obtained one or more values.

11. The computer-implemented method according to any of claims 1 to 10, wherein the CNN comprises a CNN architecture, and wherein the CNN architecture comprises a linear activation node as a final node.

12. A data processing apparatus (600) for estimating seed cotton yield, wherein the data processing apparatus comprises a processor being configured to carry out the method according to any of claims 1 to 11.

13. A computer program product comprising instructions which, when the program is executed by a computing system, cause the computing system to carry out the method according to any of claims 1 to 11.

14. A computer-readable medium having stored thereon the computer program product according to claim 13.

15. Use of the method according to any of claims 1 to 11 for at least one of: determining when to harvest the seed cotton breeding plots, determining a harvesting schedule for one or more of the seed cotton breeding plots, outputting, to a harvester and / or control device for controlling a harvester and / or to a user an indication when to harvest the seed cotton breeding plots on a field and / or the harvesting schedule, outputting selection support data for making a selection decision in breeding and / or trait development, and outputting grouping support data for grouping seed cotton breeding plots in the field in accordance with seed cotton breeding plot fitness.

Images