Kmeans clustering training

This function takes in a collection of training images and fits a patch-based kmeans cluster model for later use in classifying cluster assignment in a target image. The target and training images may be in grayscale or RGB image format.

plantcv.learn.train_kmeans(img_dir, k, out_path="./kmeansout.fit", prefix="", patch_size=10, sigma=5, sampling=None, seed=1, num_imgs=0, n_init=10)

outputs A model fit file

  • Parameters:

    • img_idr = Path to directory where training images are stored
    • k = Number of clusters to fit
    • out_path = Path to directory where the model output should be stored
    • prefix = Keyword for target images. Anything in img_dir without the prefix will be skipped
    • patch_size = Size of the NxN neighborhood around each pixel
    • sigma = Gaussian blur sigma. Denotes severity of gaussian blur performed before patch identification
    • sampling = Fraction of image from which patches are identified
    • seed = Seed for determinism of random elements like sampling of patches
    • num_imgs = Number of images to use for training. Default is all of them in img_dir with prefix
    • n_init = Number of random initiations tried by MiniBatchKMeans. The algorithm is run on the best one
  • Context:

    • Used to fit a kmeans cluster model on a set of training images. Intended to be used with pcv.predict_kmeans and pcv.mask_kmeans downstream, which are documented here.
  • Example use:

    • Below

from plantcv import plantcv as pcv


# Use 10 images to train 6 clusters with a patch size of 4
pcv.learn.train_kmeans(img_dir="./silphium_integrifolium_root_images", 
             out_path="./kmeansout_.fit", prefix="Silphium", k=6, patch_size=4, num_imgs=10)

Source Code: Here