RIVeR-Lab / hyperspectral_macro_plastic_detection

Macroplastic detection in aquatic environments with hyperspectral imaging

Hyperspectral Macroplastic Detection

Code organization

Processing

Contains repos to process raw data into reflectance calibrated and registered images

Experiments

plastics_dry - Reference plastic tiles in black plastic container

plastics_dry_pure_sand - Reference plastic tiles in dry sand river bed

plastics_dry_wild_sand - Real world containers in dry sand river bed

plastics_teaser - Plastic tiles scattered in real sand environment

plastics_wet_pure - Plastic reference tiles submerged under an 5 cm of water (if tiles were less bouyant than water)

plastics_wet_wild_settled - Real world containers in wet river bed with settled water

plastics_wet_wild_turbid - Real world containers in wet river bed with turbid, cloudy water

Calibration

Contains homographies and dark/white reference images to normalize images, and mask to extract usable overlap range.

Loading data

Cube

Hyperspectral data is stored as a compressed numpy object for efficiency.

import numpy as np

# This loads a dictionary
data = np.load(<<PATH to npz file>>)
# Extract the 3D datacube
cube = data['cube']

RGB Image

import cv2

img = cv2.cvtColor(cv2.imread(<<PATH to png file>>), cv2.COLOR_BGR2RGB)

Labels

import cv2

img = cv2.imread(<<PATH to png file>>), cv2.IMREAD_UNCHANGED)