Method of detecting and quantifying sun-drying crops using satellite derived spectral signals

Abstract

The unique methodology and utility seeking patient protection in this application is using satellite imagery to quantify and derive harvest statistics of sun-drying crops. High resolution imagery can be used to geospatially define known coffee drying basins however it is impractical for continuous observation as it is costly and infrequent. The geospatially defined regions of interests of coffee drying basins can be matched with more frequent, lower resolution, multispectral satellite imagery (such as Sentinel-2). The signals can be tested against the known spectral signatures of washed and unwashed coffee to determine whether or not each pixel contains coffee. The result will yield a classified region of interest which can be used to determine the quantity of drying coffee and the washed to unwashed ratio of a harvest. With regular monitoring across multiple temporal scenes the harvest's seasonality and historical change can be derived.

Claims

1. A method of assessing a harvest of sun drying crops, comprising: capturing a multispectral satellite or an aerial image of a coffee estate sun-drying beans including areas where coffee is sun-dried; defining a geospatial region of interest within the image; calculating a Normalized Difference Vegetation Index and a Normalized Difference Water Index for each pixel within the geospatial region of interest; testing each pixel within the geospatial region of interest for a reflectance value matching at least one spectral signature including washed or unwashed coffee dries; classifying each pixel within the geospatial region of interest based on the spectral signature and the calculated Normalized Difference Vegetation Index and the Normalized Difference Water Index; comparing temporal changes of the Normalized Difference Vegetation Index and the Normalized Difference Water Index for each classified pixel within the geospatial region of interest; analyzing a harvest cycle and an amount of crop drying based on the spectral signature, the Normalized Difference Vegetation Index and the Normalized Difference Water Index for each of the classified pixel within the geospatial region of interest; counting the classified pixels to monitor and quantify the amount of crop dying and the harvest cycle; assessing the coffee estate sun-drying beans after monitoring and quantifying the amount of crop dying and the harvest cycle; and deriving a regional or global coffee productivity levels using the assessing of the coffee estate sun drying beans and comparing the productivity levels across a sample of farms.

2. The method according to claim 1 comprising: testing and classifying each pixel against a correlation of decreasing Normalized Difference water Index and increasing Normalized Difference Vegetation Index as containing unwashed beans drying.

3. The method of claim 2, wherein the classification of each pixel includes areas both inside and outside of the geospatial region of interest.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 illustrates a high-resolution satellite image rendered in visual light over a coffee estate sun-drying beans.

(2) FIG. 2 illustrates defining a region of interest over an area where beans are sun-dried.

(3) FIG. 3 shows the region of interest cropped and delineated to represent the pixilation of a lower resolution satellite image.

(4) FIG. 4 shows the classified region of interest.

(5) FIG. 5 shows the cross-section of a coffee cherry.

(6) FIG. 6 shows a drying unwashed coffee cherry's Normalized Difference Water Index to Normalized Difference Vegetation Index relationship.

(7) FIG. 7 shows a flowchart of the classification process.

(8) TABLE-US-00002 Drawings - Reference Numerals 10 unwashed coffee 12 washed coffee 14 surrounding forest 16 cleared land 18 polygon edge 20 polygon defined region of interest 22 polygon border 24 one low-resolution pixel 26 classified washed coffee 28 classified unwashed coffee 30 classified no coffee 32 cropped pixel (bare basin) 34 coffee bean 36 silverskin and parchment of a coffee cherry 38 pulp of a coffee cherry 40 sample points of sun drying unwashed coffee 42 best fit line

DETAILED DESCRIPTION OF SPECIFIC EMBODIMENTS

(9) The present invention is a method of taking signals captured from an earth orbiting satellite and deriving the physical quantity of a harvested sun-drying crop within a defined region of interest. The present invention takes advantage of the remote sensing of visible, near-infrared, and short-wave radiation reflected from a region of interest in order to generate the initial raw data. The raw data is then converted to two vegetation indices (NDVI and NDWI). The indices and raw data is then classified.

(10) While the present invention is described herein with reference to illustrative embodiments for particular applications for analyzing sun-drying coffee beans, it should be understood that the invention is not limited thereto. Those having ordinary skill in the art and access to the teachings provided herein will recognize additional modifications, applications, and embodiments within the scope thereof and additional fields in which the present invention would be of significant utility. This would most notably be recognized as modifying the methodology to detect for other sun-drying crops such as cacao and pepper.

(11) The derived classification equations can be altered and adapted to detect for the presence of other sun-drying crops within a region of interest. The region of interest may be expanded beyond areas of known sun-drying activities as a means of mapping new areas of sun-drying activity.

(12) The linear regression model used for classifying drying unwashed coffee beans may also be applied to other sun-dying fruits or beans and used as a means of determining dryness.

(13) Overview of Remote Sensing in Agriculture

(14) Technological advances in remote sensing, notably satellite image acquisition, has enabled new methods for quantifying, estimating, and assessing the health of and quantity of agricultural harvests on a global scale. Some of the data used is high-resolution visual imagery. High-resolution satellite imagery is generally considered to have a resolution at three squared meters or less per pixel. The image is rendered from three bands containing reflectance values of each of the primary colors. This imagery can be used for counting ships passing through a port, for example, a sign of economic activity.

(15) Multispectral satellites collect light not only in the visual wavelengths, but also in wavelengths outside of the visual spectrum such as near-infrared, shortwave-infrared, thermal, and others. With multispectral imagery more complex analytics can be derived. The Normalized Difference Vegetation Index is calculated using an equation factoring in the visual red and the near-infrared bands of an image. NDVI is useful in detecting healthy live vegetation. This can be compared to historical images and harvest statistics to derive harvest yield change in sun-grown crops.

(16) Overview of Coffee as a Shade-Grown Crop

(17) Unlike corn, coffee is often grown under a forested canopy making direct detection of the plant unfeasible. Once picked however, coffee is sun dried for several days.

(18) FIG. 5 shows the cross section of a coffee cherry. As coffee is a fruit, with just the inner seed (FIG. 5-34) becoming what we use to brew coffee, the fleshly fruit covering, or pulp. (FIG. 5-38) of the coffee is either first removed in a process known as washing or the entire cherry is dried without removing the pulp. The silverskin coating of the bean (FIG. 5-36) is not removed before either drying process.

(19) During harvest season coffee farmers will sun-dry washed, unwashed, or a combination. As the washing process removes the pulp, the cherry's physical properties are altered. Washed and unwashed coffee have two unique spectral signatures.

(20) Temporal Variation in Spectral Curves

(21) Many objects have a spectral signature that is dynamic over time. An agricultural field starts as bare soil. As vegetation begins to emerge and fill the field there is a lowering of the red reflectance (due to increased chlorophyll) and an increase in near-infrared reflectance (due to increased cellular structure). The spectral curve of this agricultural site thus begins to take the signature of healthy green vegetation.

(22) The dynamic nature of spectral curves also exists in the context of sun-drying beans. As unwashed coffee cherries dry, the water content decreases (decreasing NDWI) and the outer pulp of the cherry's cellular structure shrivels, increasing chlorophyll density (and increasing NDVI).

First EmbodimentDefining Region of Interest

(23) One embodiment of the procedure is illustrated in FIG. 1 and FIG. 2. FIG. 1 illustrates visual high resolution satellite imagery of a coffee estate. In the illustration one can see the dense forest 14 under which coffee is shade grown. A clearing in the forest 16 provides the necessary sun exposure to allow the coffee to sun-dry. In the drying area, during harvest season, one may observe both washed 12 and unwashed 10 coffee. In high-resolution imagery these two are visually distinct. Unwashed coffee 10 is darker than washed coffee 12.

(24) Image acquisition at high resolution is expensive and temporally infrequent, limiting it as a practical means of continuous observation. It is useful however in defining regions of interests.

(25) FIG. 2. Shows how a georeferenced region of interest is created by selecting points 18 along the edge of the drying basin, following its perimeter 22, to create a polygon 20 of the area in which coffee is sun-dried. From high resolution imagery one can determine the expanse of the region 22. The drying basin is often built of cement or packed dirt, distinct from its surroundings.

(26) With a georeferenced region of interest defined 22, the polygon may be exported as a shapefile and the signals of a multispectral satellite image within may be assessed and classified.

(27) Regions of interest can also be created by other means, such as recording ground points along a drying basin's perimeter 22 and manually building the polygon with GIS software.

Second EmbodimentRegion of Interest Overlay

(28) Low-resolution multispectral satellite imagery (at a resolution of 10 meters squared or more per pixel) is useful for monitoring harvest yields on a regular temporal scale. The resolution is sufficient for classification and the temporal scale frequent. The signals collected by the Sentinel 2 satellite contains the necessary frequencies outside of the visual light spectrum essential for calculating NDVI and NDWI.

(29) The geospatially defined region of interest (FIG. 2.-22) can be used as an overlay to map which pixels are to be classified.

(30) FIG. 3 shows the same region of interest 20 illustrated to the pixel resolution quality of a low-resolution multispectral satellite such as Sentinel-2 which captures multispectral imagery at a 10 meter squared resolution. In FIG. 3 each pixel 24 can be seen delineated as a square. For each pixel 24 the satellite obtains wavelength signals across the visual, near-infrared, and short-wave infrared frequencies.

Third EmbodimentClassification

(31) For each pixel within the defined region of interest 20 a classification test is carried out using bands inside and out of the visual spectrum. The wavelengths collected by Sentinel-2 are broken into the following bands:

(32) TABLE-US-00003 Sentinel-2 Bands Central Wavelength (m) B2Blue 0.490 B3Green 0.560 B4Red 0.665 B5Vegetation Red Edge 0.705 B6Vegetation Red Edge 0.740 B7Vegetation Red Edge 0.783 B8NIR 0.842 B11SWIR 1.610

(33) From these bands, first NDWI and NDVI are calculated:
NDWI=(Near Infrared Band 8Short Wave Infrared Band 11)/(Near Infrared Band 8+Short Wave Infrared Band 11)
NDVI=(Near Infrared Band 8Red Band 4)/(Near Infrared Band 8+Red Band 4)

(34) NDVI and NDWI are amended to the dataset of each pixel. Tests are then performed to determine if the pixel contains either washed coffee, unwashed coffee, cloud coverage, bare drying basin (no coffee), or unknown substances.

(35) Each pixel is compared to the spectral signature of unwashed coffee to test for the presence of unwashed coffee.

(36) Is the recorded reflectance of the blue band (B2) greater than 31% the total of all visual bands?
Is: B2/(B2+B3+B4)>31%

(37) Is the reflectance of blue greater than the reflectance of green?
Is: B2>B3

(38) Is near-infrared reflectance greater than visual red reflectance?
Is: B8>B4

(39) Lastly, does the relationship between the pixel's NDVI and NDWI fall close to the defined regression illustrated in FIG. 6.-42. The margin of error can be adjusted by altering the b values of the equation (y=m*x+b) relative to standard deviation.
Is: NDWI>1.5*NDVI+0.55
and
Is: NDWI<1.5*NDVI+0.75

(40) The pixel is classified as containing sun-drying unwashed coffee if it meets these tests.

(41) In testing for the presence of washed coffee, a classification model testing the pixel's spectral signature is used. As the outer pulp of the coffee cherry 38 is already removed before drying, one cannot rely on the same NDVI to NDWI relationship as employed for testing unwashed coffee. In mapping a pixel as washed, the following tests are employed.

(42) First NDVI is tested. As the bean is vegetative matter with chlorophyll reflecting high levels of near-infrared and low levels of the visual red light.
Is NDVI>0.1

(43) NDWI of washed coffee will be less than zero as the pulp 38 has already been removed. Without this more short-wave infrared and less near-infrared is reflected.
Is NDWI<0

(44) Visual bands are then tested using unwashed coffee's known spectral signature:

(45) Is red reflectance greater than 36% of the combined visual bands?
Is B4 Red/(B2 Blue+B3 Green+B4 Red)>36%

(46) Is the red reflectance values greater than both that of blue and green?
Is B4 Red>B3 Green?
and
Is B4 Red>B2 Blue?

(47) Washed coffee also reflects greater values of green than blue, so:
Is B3 Green>B2 Blue?

(48) It is also important to test for cloud coverage which may obscure classification within the drying basin. In Sentinel-2 imagery, if cirrus or opaque clouds are present, a band called QA60 is amended to the dataset. To test for these clouds, the following suffices:
Is QA60>100.

(49) If the test returns true, the pixel is classified as cloud.

(50) A more complex cloud mask incorporating several bands can also be used:
Is (NDVI<0.4) AND ((B8/B3)<1.2) AND ((B2/B11)>0.70))
or
Is (B8A<1050) AND (B8<1200) AND (B3<2000) AND (NDVI<0.10))
or
Is ((B3+B8A)>40000) AND (B3<B2) AND (B4<B2))
or
Is (QA60>25)
or
Is (B8A>4500)

(51) The if/or statements allow for the testing of different types of clouds as different clouds have different spectral signatures based on their altitude, contents, and character.

(52) Should each pixel classify as neither washed coffee, unwashed coffee, or cloud, it is tested for any traces of vegetative matter.
Is NDVI<0.09

(53) If this test is true, it is confident to say that the basin contains little to no vegetative matter.

(54) If this test fails however, the pixel is classified as unknown. Unknown pixels may be areas containing traces of drying coffee, but perhaps not enough to match with washed or unwashed coffee's spectral signature. Otherwise these pixels may be areas containing vegetative matter other than coffee.

(55) Operation

(56) Using the scene from FIG. 1 one could expect a classification as illustrated in FIG. 4. Again the washed coffee 10 is classified as washed coffee 26, the unwashed coffee 12 as unwashed coffee 28, and the empty portions of the drying basin are classified as empty 30.

(57) Some pixels will cover areas both inside and outside of the defined region of interest. These pixels are cropped. FIG. 4 illustrates cropped pixels 32. These pixels will contribute as fractions relative to their area within the region of interest 20 when calculating total area of drying coffee.

CONCLUSION, RAMIFICATIONS, AND SCOPE

(58) The area of the resulting classification is totaled. Statistics are calculated such as the percentage of washed and unwashed coffee within a basin. Coffee yield figures are accurately derived based on the surface expanse of the dying coffee. When coffee dries the depth of the drying coffee is consistently kept at a minimal depth of just one bean to avoid rotting. Volume is thus easily calculated.

(59) This method allows for the monitoring of coffee farms globally. Running this procedure across a sample of farms will be a strong indicator of regional or global coffee productivity levels.

(60) This method is not limited to coffee. Any sun drying bean, fruit, or other vegetative matter may be classified and monitored simply by altering the classification algorithm according to the properties of the substance in question.

(61) This method has been developed to a high level of usability. Satellite derived coffee harvest analytics have already been supplied to world leading green coffee service groups. Site specific analytics have also been back tested against multi-year ground data. Lastly, regional Brazilian yield figures have been accurately calculated using a database of sample Brazilian sample sites.