Extract lines of rice terraces from satellite image

Question

I have a Pleiades image and multiple rice field boundaries in a shape file. What I want to do is for every feature in the shape file, clip the Pleiades image and detect the lines of the field terraces. I have already managed to detect the lines using the difference of gaussian blurs, but they look quite messy. Is there a way to do this smarter with better results?

This is the part of my script that detects the lines:

from skimage import filters, exposure

 # I average all bands to a single band
avgs = [ avg_bands(i) for i in clips]


result_rasters = []

# normalize images between 0 and 255
# using scikit-image library for gauss and median filter
for raster in avgs:
    dif_gauss = normalize(filters.difference_of_gaussians(
    raster,
    low_sigma=0,
    high_sigma=3),255)

    med_dg = filters.median(dif_gauss)

    # I look for values smaller than the mean of the image
    med_dg = med_dg < med_dg.mean() 

    result_rasters.append(med_dg)