Image segmentation for improved consistency in image-interpretation of opium poppy

Abstract

The image-interpretation of opium poppy crops from very high resolution satellite imagery forms part of the annual Afghanistan opium surveys conducted by the United Nations Office on Drugs and Crime and the United States Government. We tested the effect of generalisation of field delineations on the final estimates of poppy cultivation using survey data from Helmand province in 2009 and an area frame sampling approach. The sample data was reinterpreted from pan-sharpened IKONOS scenes using two increasing levels of generalisation consistent with observed practice. Samples were also generated from manual labelling of image segmentation and from a digital object classification. Generalisation was found to bias the cultivation estimate between 6.6% and 13.9%, which is greater than the sample error for the highest level. Object classification of image-segmented samples increased the cultivation estimate by 30.2% because of systematic labelling error. Manual labelling of imagesegmented samples gave a similar estimate to the original interpretation. The research demonstrates that small changes in poppy interpretation can result in systematic differences in final estimates that are not included within confidence intervals. Segmented parcels were similar to manually digitised fields and could provide increased consistency in field delineation at a reduced cost. The results are significant for Afghanistan's opium monitoring programmes and other surveys where sample data are collected by remote sensing.

Opium Poppy Detection Using Deep Learning

“…For a region with a fixed area, the number of image pixels varies with the spatial resolution, which can be approximately calculated by Formula (6), where Numbers represents the total number of pixels, W and H are the region width and length, respectively, and Res represents the image spatial resolution. When setting the overlap (default = 150) during prediction, the total time required can be calculated by Formula (7), where Speed is the time required to detect each patch image:…”

“…and H are the region width and length, respectively, and Res represents the image spatial resolution. When setting the overlap (default = 150) during prediction, the total time required can be calculated by Formula (7), where Speed is the time required to detect each patch image:…”

“…The UNODC uses a statistical sample method combined with remote sensing images to estimate poppy cultivation in monitoring areas. The existing methods [4,6,7] focus almost entirely on the total planting acreage in some provinces and countries. The total planting acreage is an important indicator in evaluating the overall planting situation; however, in practice, obtaining poppy parcel location information is more important for eradicating poppies [5].…”

“…Some studies have proposed different methods to improve the poppy detection accuracy or efficiency of remote sensing image-based techniques. Simms et al used imagery-based stratification [6] and image segmentation [7] to improve opium cultivation estimates in Afghanistan, but estimates were still based on sampling methods. Jia et al [8] demonstrated that opium poppies could be distinguished from coexisting crops in many surveyed wavebands using a field survey spectrum at the canopy level on an official poppy-growing region in China.…”

Διερεύνηση Τεχνικών Τηλεπισκόπησης Και Ανάπτυξη Μεθοδολογίας Για Την Ακριβή Εκτίμηση Της Καύσιμης Ύλης Με Βάση Τη Βλάστηση Χρησιμοποιώντας Δύο Εποχικές Δορυφορικές Εικόνες Υψηλής Ανάλυσης

“…Πηγές σφαλμάτων στην ερμηνεία εικόνας είναι η λανθασμένη επισήμανση τμημάτων γης και ανακρίβειας των ορισμών της τάξης, για παράδειγμα η τοποθέτηση ενός ορίου σε ένα συνεχές (Foody, 2002 Παραδείγματα εφαρμογής της μεθόδου χειροκίνητης ταξινόμησης έπειτα από εφαρμογή τεχνικής κατάτμησης εικόνας, μπορεί κανείς να εντοπίσει στην βιβλιογραφία, όπως για παράδειγμα (Lathrop et al, 2006;Kampouraki et al, 2008;Lang et al, 2009;Simms et al, 2016 πολλούς ερευνητές έχει αναφερθεί ότι η μείωση του φασματικού θορύβου έπειτα από τη συσσωμάτωση των εικονοστοιχείων σε αντικείμενα δε βελτιώνει απαραίτητα και τα αποτελέσματα της ταξινόμησης π.χ. (Koch et al, 2003;Μαλλίνης, 2006;Gao & Mas, 2008;Fung et al, 2008;Petropoulos et al, 2012;Jebur et al, 2013;Jonsson, 2015;Goodin et al, 2015;Trang et al, 2016), ενώ έχουν επισημανθεί οι αδυναμίες εφαρμογής των αντικειμενοστραφών προσεγγίσεων όσον αφορά στην ταξινόμηση δασικών ειδών, κυρίως λόγω μη επιθυμητής συσσωμάτωσης σε αντικείμενα τύπων δασών με διαφορετική πρόσμιξη ειδών.…”