Method and apparatus for automated plant necrosis
A method of real-time plant selection and removal from a plant field including capturing a first image of a first section of the plant field, segmenting the first image into regions indicative of individual plants within the first section, selecting the optimal plants for retention from the first image based on the first image and the previously thinned plant field sections, sending instructions to the plant removal mechanism for removal of the plants corresponding to the unselected regions of the first image from the second section before the machine passes the unselected regions, and repeating the aforementioned steps for a second section of the plant field adjacent the first section in the direction of machine travel.
1 . A method for a farming machine to treat a plant in a field comprising a plurality of plants, the method comprising:
accessing an image of the field comprising the plurality of plants, the image captured as the farming machine moves through the field on a current pass;
segmenting the image into a plurality of field regions by applying a machine learning model to the image, the machine learning model processing the image to localize the plurality of plants within the plurality of field regions, with each of the field regions localizing a plant of the plurality of plants within a two dimensional area in the image;
selecting, for each field region of the plurality of field regions, a spray treatment for the plant localized in the field region based on characteristics describing the localized plant and parameters describing a plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field; and
sending treatment instructions for each region of the plurality of field regions to a plurality of spray mechanisms of the farming machine, the treatment instructions indicating the spray treatment to apply to each region before the farming machine travels past the region in the field on the current pass.
2 . The method of claim 1 , wherein applying the machine learning model to localize the plant within the two dimensional image comprises:
classifying pixels in the image as points representing the plant; and
determining a position of the points representing the plant.
3 . The method of claim 2 , wherein classifying pixels in the image as points representing the plant comprises:
for each pixel in the image:
determining a confidence the pixel is a point of interest representing the plant; and
determining, based on the confidence, the point of interest represents the plant.
4 . The method of claim 2 , wherein the method further comprises:
segmenting the image to localize the pixels classified as representing the plant at the determined position, the localized pixels within a portion of the two dimensional image.
5 . The method of claim 1 , wherein the previous farming pass occurs during a first period of time and the current farming pass occurs during a second period of time after the first period of time.
6 . The method of claim 1 , wherein selecting the spray treatment for the localized plant comprises:
determining, based on information from the previous pass, if a first substance or a second substance is suitable for removing the localized plant from the field; and
selecting the first substance or the second substance as the spray treatment based on the determination.
7 . The method of claim 1 , wherein parameters describing the plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field comprises a yield metric for plants treated on the previous pass of the farming machine through the field.
8 . The method of claim 1 , wherein parameters describing the plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field comprises a size of plants treated on the previous pass of the farming machine through the field.
9 . The method of claim 8 , wherein parameters describing the size of plants treated on the previous pass are compared to an expected size of plants in the field.
10 . The method of claim 1 , wherein parameters describing the plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field comprises a shape of plants treated on the previous pass of the farming machine through the field.
11 . The method of claim 10 , wherein parameters describing the shape of plants treated on the previous pass are compared to an expected shape of plants in the field.
12 . A farming machine:
a camera configured to capture images of a field comprising a plurality of plants as the farming machine travels through the field;
a plurality of spray treatment mechanism;
one or more processors; and
a non-transitory computer readable medium storing instructions that, when executed, cause the one or more processors to:
access an image of the field comprising the plurality of plants, the image captured as the farming machine moves through the field on a current pass;
segment the image into a plurality of field regions by applying a machine learning model to the image, the machine learning model processing the image to localize the plurality of plants within the plurality of field regions, with each of the field regions localizing a plant of the plurality of plants within a two dimensional area in the image;
select, for each field region of the plurality of field regions, a spray treatment for the plant localized in the field region based on characteristics describing the localized plant and parameters describing a plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field; and
send treatment instructions for each region of the plurality of field regions to the plurality of spray mechanisms of the farming machine, the treatment instructions indicating the spray treatment to apply to each region before the farming machine travels past the region in the field on the current pass.
13 . The farming machine of claim 12 , wherein applying the machine learning model to localize the plant within the two dimensional image further causes the one or more processors to:
classify pixels in the image as points representing the plant; and
determine a position of the points representing the plant.
14 . The farming machine of claim 13 , wherein classifying pixels in the image as points representing the plant further causes the one or more processors to:
for each pixel in the image:
determine a confidence the pixel is a point of interest representing the plant; and
determine, based on the confidence, the point of interest represents the plant; and
wherein executing the instructions further cause the one or more processors to:
segment the image to localize the pixels classified as representing the plant at the determined position, the localized pixels within a portion of the two dimensional image.
15 . The farming machine of claim 12 , wherein the previous farming pass occurs during a first period of time and the current farming pass occurs during a second period of time after the first period of time.
16 . The farming machine of claim 12 , wherein parameters describing the plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field comprises a yield metric for plants treated on the previous pass of the farming machine through the field.
17 . The farming machine of claim 12 , wherein parameters describing the plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field comprises a size and a shape of plants treated on the previous pass of the farming machine through the field.
18 . A non-transitory computer readable medium storing instructions for a farming machine to treat a plant in a field comprising a plurality of plants, the instructions when executed by one or more processors, causing the one or more processors to:
accessing an image of the field comprising the plurality of plants, the image captured as the farming machine moves through the field on a current pass;
segmenting the image into a plurality of field regions by applying a machine learning model to the image, the machine learning model processing the image to localize the plurality of plants within the plurality of field regions, with each of the field regions localizing a plant of the plurality of plants within a two-dimensional area in the image;
selecting, for each field region of the plurality of field regions, a spray treatment for the plant localized in the field region based on characteristics describing the localized plant and parameters describing a plurality of prior plants treated by the farming machine on a previous pass of the farming machine through the field; and
sending treatment instructions for each region of the plurality of field regions to a plurality of spray mechanisms of the farming machine, the treatment instructions indicating the spray treatment to apply to each region before the farming machine travels past the region in the field on the current pass.