The use of cloud based imagery services for understanding landscape change in support of opium monitoring in Afghanistan
Lead Research Organisation:
Cranfield University
Department Name: School of Water, Energy and Environment
Abstract
Information on illicit poppy cultivation in Afghanistan is of critical importance to the opium monitoring programme of the United Nations Office on Drugs and Crime (UNODC). The pattern of cultivation is constantly evolving because of environmental pressures, such as water availability, and social and economic drivers related to counter narcotics activity. Remote sensing already plays a key role in gathering information on the area of opium cultivation and its spatial distribution. The shift to cloud computing opens up exciting possibilities for extracting new information from the huge amounts of satellite data from long-term earth observation programmes. You will test the hypothesis that inter-annual and intra-seasonal changes in vegetation growth cycles are predictors of poppy cultivation risk. This will involve using emerging cloud based technologies for processing image data into accurate and timely information on vegetation dynamics relating to opium cultivation.
Organisations
Studentship Projects
Project Reference | Relationship | Related To | Start | End | Student Name |
---|---|---|---|---|---|
NE/M009009/1 | 05/10/2015 | 31/12/2022 | |||
1948488 | Studentship | NE/M009009/1 | 25/09/2017 | 01/04/2021 | Alex Hamer |
Description | - The training rationale for pixelwise classification of agricultural land using a deep learning technique called 'Convolutional Neural Networks' (CNNs) and satellite imagery has been identified with different training configurations. This will optimise the use of labelled data to delineate agriculture land using this technique. - Continual year-on-year updating of the CNN model led to greater classification accuracy with less data required for each additional year added as training for agricultural land delineation across Helmand and Kandahar, Afghanistan (94.57% overall accuracy). This model can be used to inform whether agriculture is present in medium resolution satellite imagery (DMC at 32 m spatial resolution) between different years. - Fully Convolutional Networks (FCNs) have been developed for extracting agricultural land across different medium resolution satellite image sensors (DMC, Landsat and Sentinel-2). This provides a sensor agnostic approach to image classification, whereas current approaches are sensor specific for classification. |
Exploitation Route | The current findings have the potential to be used in an operational capacity by implementing an appropriate training strategy for both CNN and FCN agricultural land delineation. The ability to train between years and improve classification accuracy whilst using less data from each subsequent year is beneficial for others attempting to carry out similar analysis, as access to labelled data is scarce to train these models. The potential to build sensor agnostic models is useful for others as it provides the opportunity to use large volumes of image data from archives of new and long-term Earth observation programmes to monitor agricultural area, particularly for deployment on cloud services such as Google Earth Engine. |
Sectors | Agriculture, Food and Drink,Environment |
Title | Convolutional Neural Network (CNN) model for delinating agricultural land from medium resolution satellite imagery |
Description | The model delineates agriculture from an input satellite image from the Disaster Monitoring Constellation (DMC) at 32m spatial resolution using training and validation data across Helmand and Kandahar province, Afghanistan between 2007 and 2009. The binary classification operates by classifying each input image chip (33 x 33 pixels) and returns either and agriculture or non-agriculture prediction class. |
Type Of Material | Computer model/algorithm |
Year Produced | 2019 |
Provided To Others? | No |
Impact | The model is able to automate the reconstruction of the agricultural area across Helmand and Kandahar province to outperform other machine learning techniques using the same approach. |
Title | Fully Convolutional Network for agricultural delineation across Helmand Province, Afghanistan from multispectral satellite imgery |
Description | The model delineates agriculture from any Level-1A input satellite image (NIR, R, G) IR-MAD normalised to a reference DMC image. The model has been trained on agriculture delineations between 2007-2017 using DMC and Landsat 8 imagery across Helmand Province, Afghanistan. The binary classification operates at any image size and returns a pixelwise output of agricultural land. |
Type Of Material | Computer model/algorithm |
Year Produced | 2020 |
Provided To Others? | No |
Impact | The model can be deployed on any satellite image using IR-MAD normalisation of the NIR, R and G spectral bands allowing monitoring of agriculture across Helmand Province, Afghanistan. |
Description | Presentation at the 63rd session of the Commission on Narcotic Drugs, United Nations Office on Drugs and Crime, Vienna |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Third sector organisations |
Results and Impact | The presentation was part of the United Nations Office on Drugs and Crime's (UNODC) side event 'Monitoring illicit crops using artificial intelligence' in Vienna, Austria. I presented part of the work I had done from this award during the session with emphasis on how deep learning using Convolutional Neural Networks (CNNs) could help with monitoring where illicit crops are being grown in Afghanistan. |
Year(s) Of Engagement Activity | 2020 |
URL | https://www.cranfield.ac.uk/press/news-2020/artificial-intelligence-could-help-with-illicit-crop-mon... |