Estimation method for crop yield

Abstract

A method for predicting crop yield: A first multispectral image is captured on the first sampling day of a first period for a first crop in a farmland; the first multispectral image is processed to obtain vegetation indicators of the sampling area in the first multispectral image; a linear regression analysis is performed on the vegetation indicators of the sampling area in the first multispectral image and the total fresh weight of the sampling area on the first harvest day to obtain a first yield prediction model; a second multispectral image is captured on the second sampling day of a second period for a second crop in a farmland; the second multispectral image is processed to obtain vegetation indicators of the sampling area in the second multispectral image; the yield of the second crop harvested on the second harvest day of the second period is predicted based on the first yield prediction model and the vegetation indicators of the sampling area in the second multispectral image.

Claims

1. A method for predicting crop yield, comprising: A first multispectral image is captured on a first sampling day within a first period for a farmland planted with a first crop, wherein an area of ​​the farmland is divided into a plurality of sampling areas, and the first period is from a first initial planting date of the first crop to a first harvest date of the first crop; an image processing device is used to process the first multispectral image to obtain a vegetation index for each of the sampling areas in the first multispectral image; a first yield prediction model is obtained by performing a linear regression analysis between the vegetation index of each of the sampling areas in the first multispectral image and a total fresh weight of each of the sampling areas on the first harvest date; a second multispectral image is captured on a second sampling day within a second period for a farmland planted with a second crop, wherein the first crop and the second crop are of the same type. The processing device performs image processing on the second multispectral image to obtain the vegetation index of each of the sampling areas in the second multispectral image; and estimates the yield of the second crop harvested on a second harvest day in the second period based on the first yield prediction model and the vegetation index of each of the sampling areas in the second multispectral image.

2. The crop yield estimation method as described in claim 1, wherein the first multispectral image and the second multispectral image are acquired by a drone equipped with a multispectral camera from the farmland.

3. The crop yield estimation method as described in claim 2, wherein processing the first multispectral image using the processing device to obtain the vegetation index for each of the sampling areas in the first multispectral image includes: The first multispectral image is divided into a plurality of grids according to a grid size; The normalized difference vegetation index is calculated for each grid of each of the sampling areas in the first multispectral image; and the normalized difference vegetation indices of each grid of each of the sampling areas in the first multispectral image are summed and divided by the number of grids in each of the sampling areas to obtain an average normalized difference vegetation index for each of the sampling areas in the first multispectral image as the vegetation index for the sampling area; wherein the first yield prediction model is obtained by performing a linear regression analysis between the average normalized difference vegetation index of each of the sampling areas in the first multispectral image and the total fresh weight of each of the sampling areas on the first harvest day.

4. The crop yield prediction method as described in claim 3, wherein the first multispectral image includes the spectra of the red band and the near-infrared band, and wherein the normalized difference vegetation index is represented by equation (1): … (1) where, NDVI is the normalized difference vegetation index for a specific region in a multispectral image, NIR is the reflectance in the near-infrared band of the specific region in the multispectral image, and RED is the reflectance in the red band of the specific region in the multispectral image, where -1 ≤ NDVI ≤ 1.

5. The crop yield estimation method as described in claim 3, wherein the first yield estimation model is represented by equation (2): Wa = 1793.48 × NDVI_1 mean + 1202.21 … (2) where, Wa is a first estimated total fresh weight of one of the sampling areas, and NDVI_1 mean is the mean normalized difference vegetation index of one of the sampling areas in the multispectral image of the farmland captured by the multispectral camera.

6. The crop yield estimation method as described in claim 2 further includes: A third multispectral image is captured by a satellite device on the first sampling day of the farmland where the first crop is grown; the third multispectral image is processed by the processing device to obtain the vegetation index of each of the sampling areas in the third multispectral image; a second yield prediction model is obtained by performing linear regression analysis on the vegetation index of each of the sampling areas in the third multispectral image and the total fresh weight of each of the sampling areas on the first harvest day; a fourth multispectral image is captured by the satellite device on the second sampling day of the second period of the farmland where the second crop is grown; the fourth multispectral image is processed by the processing device to obtain the vegetation index of each of the sampling areas in the fourth multispectral image; and the yield of the second crop harvested on the farmland on the second harvest day of the second period is predicted based on the second yield prediction model and the vegetation index of each of the sampling areas in the fourth multispectral image.

7. The crop yield estimation method as described in claim 6, wherein the image processing of the third multispectral image using the processing device to obtain the vegetation index for each of the sampling areas in the third multispectral image includes: The third multispectral image is divided into a plurality of grids according to a grid size; According to Equation (1), a normalized difference vegetation index is calculated for each grid of each of the sampling areas in the third multispectral image; and the normalized difference vegetation indices of each grid of each of the sampling areas in the third multispectral image are summed and divided by the number of grids in each of the sampling areas to obtain an average normalized difference vegetation index for each of the sampling areas in the third multispectral image as the vegetation index for the sampling area; wherein the second yield prediction model is obtained by performing a linear regression analysis between the average normalized difference vegetation index of each of the sampling areas in the third multispectral image and the total fresh weight of each of the sampling areas on the first harvest day.

8. The crop yield estimation method as described in claim 7, wherein the third multispectral image includes spectra in the red and near-infrared bands, and wherein the normalized difference vegetation index is represented by equation (1): … (1) where, NDVI is the normalized difference vegetation index for a specific region in a multispectral image, NIR is the reflectance in the near-infrared band of the specific region in the multispectral image, and RED is the reflectance in the red band of the specific region in the multispectral image, where -1 ≤ NDVI ≤ 1.

9. The crop yield estimation method as described in claim 7, wherein the second yield estimation model is represented by equation (3): Wb = 2610.31 × NDVI_2 mean + 413.92 … (3) where, Wb is a second estimated total fresh weight of one of the sampling areas, and NDVI_2 mean is the mean normalized difference vegetation index of one of the sampling areas in the multispectral image of the farmland captured by the satellite device.

10. The crop yield estimation method as described in claim 9 further includes: The processing device performs linear regression analysis on the first yield prediction model and the second yield prediction model to obtain a third yield prediction model, which is represented by equation (4): Wc = 0.6871 × Wb + 917.82 … (4) where Wc is a third estimated total fresh weight of one of the sampling areas. The crop yield prediction method further includes: predicting the yield of the second crop harvested on the farmland on the second harvest day of the second period based on the third yield prediction model and the average normalized difference vegetation index of the second yield prediction model and each of the sampling areas in the fourth multispectral image.

11. The crop yield estimation method as described in claim 5, 9 or 10, wherein the first crop and the second crop are of the type of rice.

12. A method for estimating crop yield as described in claim 4 or 8, wherein the grid size ranges from (0.85 m x 0.85 m) to (1 m x 1 m).

13. The crop yield estimation method as claimed in claim 1, wherein the first sampling day is one of the ninetieth to one hundredth days from the first initial day of the first period; and the second sampling day is one of the ninetieth to one hundredth days from the second initial day of the second period when the second crop is planted.

14. A method for predicting crop yield, comprising: A first visible light image is captured on a first sampling day within a first period for a farmland planted with a first crop, wherein an area of ​​the farmland is divided into a plurality of sampling areas, and the first period is from a first initial planting date of the first crop to a first harvest date of the first crop; an image processing device is used to process the first visible spectral image to obtain a vegetation index for each of the sampling areas in the first visible spectral image; a fourth yield prediction model is obtained by performing a linear regression analysis between the vegetation index of each of the sampling areas in the first visible spectral image and a total fresh weight of each of the sampling areas on the first harvest date; a second visible spectral image is captured on a second sampling day within a second period for a farmland planted with a second crop, wherein the first crop and the second crop are of the same type. The processing device performs image processing on the second visible spectral image to obtain the vegetation index of each of the sampling areas in the second visible spectral image; and estimates the yield of the second crop harvested on the farmland on a second harvest day during the second period based on the fourth yield prediction model and the vegetation index of each of the sampling areas in the second visible spectral image.

15. The crop yield estimation method as described in claim 14, wherein processing the first visible spectral image using the processing apparatus to obtain the vegetation index for each of the sampling regions in the first visible spectral image includes: The first visible spectrum image is divided into a plurality of grids according to a grid size; Calculate a normalized difference vegetation index (NDVI) for each grid in each of the sampling regions in the first visible spectral image; and sum the NDVIs for each grid in each of the sampling regions in the first visible spectral image and divide by the number of grids in each sampling region to obtain an average NDVI for each sampling region in the first visible spectral image as the vegetation index for that sampling region; wherein the fourth yield prediction model is obtained by performing a linear regression analysis between the average NDVI for each sampling region in the first visible spectral image and the total fresh weight of each sampling region on the first harvest day.

16. The crop yield estimation method as described in claim 15, wherein the fourth yield estimation model is represented by equation (5): Wd = 18559.5 × vNDVI mean – 1959.1 … (5) where, Wd is the fourth estimated total fresh weight of one of the sampling areas, and vNDVI mean is the mean normalized difference in visible light vegetation index of one of the sampling areas.

17. The crop yield estimation method as described in claim 14, wherein the image processing of the first visible spectral image using the processing apparatus to obtain the vegetation index for each of the sampling regions in the first visible spectral image includes: The first visible spectrum image is divided into a plurality of grids according to a grid size; The process involves calculating a normalized difference vegetation index (NDVI) for each grid in each of the sampling regions of the first visible spectral image; converting the NDVI for each grid in each of the sampling regions of the first visible spectral image into a normalized difference vegetation index (NDI) using a convolutional neural network; summing the NDIs for each grid in each of the sampling regions of the first visible spectral image and dividing by the number of grids in each sampling region to obtain an average normalized difference vegetation index for each sampling region in the first visible spectral image, which is then used as the vegetation index for that sampling region; wherein the fourth yield prediction model is obtained by performing a linear regression analysis between the average normalized difference vegetation index for each sampling region in the first visible spectral image and the total fresh weight of each sampling region on the first harvest day.

18. The crop yield estimation method as described in claim 16, wherein the first crop and the second crop are of the type of rice.

19. A method for estimating crop yield as described in claim 15 or 17, wherein the grid size ranges from (0.85 m x 0.85 m) to (1 m x 1 m).

20. The crop yield estimation method as described in claim 14, wherein the first sampling day is one of the ninetieth to one hundredth days from the first initial day of the first period; and the second sampling day is one of the ninetieth to one hundredth days from the second initial day of the second period when the second crop is planted.

Images