US 12,307,645 B2
Quantifying biotic damage on plants, by separating plant-images and subsequently operating a convolutional neural network
Aranzazu Bereciartua-Perez, Derio (ES); Artzai Picon Ruiz, Derio (ES); Corinna Maria Spangler, Ludwigshafen (DE); Christian Klukas, Limburgerhof (DE); Till Eggers, Ludwigshafen (DE); Ramon Navarra-Mestre, Limburgerhof (DE); and Jone Echazarra Huguet, Derio (ES)
Assigned to BASF SE, Ludwigshafen am Rhein (DE)
Appl. No. 17/802,590
Filed by BASF SE, Ludwigshafen am Rhein (DE)
PCT Filed Mar. 5, 2021, PCT No. PCT/EP2021/055645
§ 371(c)(1), (2) Date Aug. 26, 2022,
PCT Pub. No. WO2021/176081, PCT Pub. Date Sep. 10, 2021.
Claims priority of application No. 20161529 (EP), filed on Mar. 6, 2020.
Prior Publication US 2023/0141945 A1, May 11, 2023
Int. Cl. G06T 7/00 (2017.01); G06Q 50/02 (2024.01); G06T 7/13 (2017.01); G06V 10/82 (2022.01)
CPC G06T 7/0004 (2013.01) [G06Q 50/02 (2013.01); G06T 7/13 (2017.01); G06V 10/82 (2022.01); G06T 2207/10024 (2013.01); G06T 2207/30188 (2013.01)] 10 Claims
OG exemplary drawing
 
1. A computer-implemented method to quantify biotic damage in leaves of crop-plants, the method comprising:
receiving a plant-image showing a crop-plant, the plant-image showing an aerial part of the crop-plant, with at least a stem, branches, and leaves and showing the ground underneath the plant, on that the plant is placed;
processing the plant-image to obtain a segmented plant-image being a contiguous set of pixels that shows a contour of the aerial part, the contour having a margin region that shows the ground partially, wherein processing the plant-image to obtain the segmented plant-image comprises:
changing color-coding of the plant-image from a first color-coding to a second color-coding, by transforming color space, the second color-coding with higher contrast between plant and ground than the first color-coding;
for the pixels in the second color-coding, differentiating the pixels of the plant-image and assigning the pixels to first and second binary values, thereby obtaining a contiguous set of pixels that are coded in the first binary value;
identifying edge pixels of the contiguous set, being pixels of the first binary value having at least one adjacent pixel of the second binary value, determining a margin line being a set of pixels in a distance to the edge pixels, wherein the distance is limited by a minimal number of pixels and a maximal number of pixels; and
replacing the pixels within the margin line by the pixels in the first color-coding; and
processing the segmented plant-image by a convolutional neural network that uses regression to obtain a damage degree, the convolutional neural network having been trained by processing damage-annotated segmented plant-images.