This page has only limited features, please log in for full access.

Dr. Do-Hyung Kim
Office of Innovation, United Nations Children’s Fund, New York, NY, USA

Basic Info


Research Keywords & Expertise

0 Machine Learning
0 Remote Sensing
0 Sustainable Development
0 Land Cover Land Use Change
0 Equitable AI

Honors and Awards

The user has no records in this section


Career Timeline

The user has no records in this section.


Short Biography

The user biography is not available.
Following
Followers
Co Authors
The list of users this user is following is empty.
Following: 0 users

Feed

Technical note
Published: 24 July 2021 in Remote Sensing
Reads 0
Downloads 0

Understanding the biases in Deep Neural Networks (DNN) based algorithms is gaining paramount importance due to its increased applications on many real-world problems. A known problem of DNN penalizing the underrepresented population could undermine the efficacy of development projects dependent on data produced using DNN-based models. In spite of this, the problems of biases in DNN for Land Use and Land Cover Classification (LULCC) have not been a subject of many studies. In this study, we explore ways to quantify biases in DNN for land use with an example of identifying school buildings in Colombia from satellite imagery. We implement a DNN-based model by fine-tuning an existing, pre-trained model for school building identification. The model achieved overall 84% accuracy. Then, we used socioeconomic covariates to analyze possible biases in the learned representation. The retrained deep neural network was used to extract visual features (embeddings) from satellite image tiles. The embeddings were clustered into four subtypes of schools, and the accuracy of the neural network model was assessed for each cluster. The distributions of various socioeconomic covariates by clusters were analyzed to identify the links between the model accuracy and the aforementioned covariates. Our results indicate that the model accuracy is lowest (57%) where the characteristics of the landscape are predominantly related to poverty and remoteness, which confirms our original assumption on the heterogeneous performances of Artificial Intelligence (AI) algorithms and their biases. Based on our findings, we identify possible sources of bias and present suggestions on how to prepare a balanced training dataset that would result in less biased AI algorithms. The framework used in our study to better understand biases in DNN models would be useful when Machine Learning (ML) techniques are adopted in lieu of ground-based data collection for international development programs. Because such programs aim to solve issues of social inequality, MLs are only applicable when they are transparent and accountable.

ACS Style

Do-Hyung Kim; Guzmán López; Diego Kiedanski; Iyke Maduako; Braulio Ríos; Alan Descoins; Naroa Zurutuza; Shilpa Arora; Christopher Fabian. Bias in Deep Neural Networks in Land Use Characterization for International Development. Remote Sensing 2021, 13, 2908 .

AMA Style

Do-Hyung Kim, Guzmán López, Diego Kiedanski, Iyke Maduako, Braulio Ríos, Alan Descoins, Naroa Zurutuza, Shilpa Arora, Christopher Fabian. Bias in Deep Neural Networks in Land Use Characterization for International Development. Remote Sensing. 2021; 13 (15):2908.

Chicago/Turabian Style

Do-Hyung Kim; Guzmán López; Diego Kiedanski; Iyke Maduako; Braulio Ríos; Alan Descoins; Naroa Zurutuza; Shilpa Arora; Christopher Fabian. 2021. "Bias in Deep Neural Networks in Land Use Characterization for International Development." Remote Sensing 13, no. 15: 2908.

Journal article
Published: 04 July 2021 in Geomatics
Reads 0
Downloads 0

Evaluation of the effectiveness of protected areas is critical for forest conservation policies and priorities. We used 30 m resolution forest cover change data from 1990 to 2010 for ~4000 protected areas to evaluate their effectiveness. Our results show that protected areas in the tropics avoided 83,500 ± 21,200 km2 of deforestation during the 2000s. Brazil’s protected areas have the largest amount of avoided deforestation at 50,000 km2. We also show the amount of international aid received by tropical countries compared to the effectiveness of protected areas. Thirty-four tropical countries received USD 42 billion during the 1990s and USD 62 billion during the 2000s in international aid for biodiversity conservation. The effectiveness of international aid was highest in Latin America, with 4.3 m2/USD, led by Brazil, while tropical Asian countries showed the lowest average effect of international aid, reaching only 0.17 m2/USD.

ACS Style

Do-Hyung Kim; Anupam Anand. Effectiveness of Protected Areas in the Pan-Tropics and International Aid for Conservation. Geomatics 2021, 1, 335 -346.

AMA Style

Do-Hyung Kim, Anupam Anand. Effectiveness of Protected Areas in the Pan-Tropics and International Aid for Conservation. Geomatics. 2021; 1 (3):335-346.

Chicago/Turabian Style

Do-Hyung Kim; Anupam Anand. 2021. "Effectiveness of Protected Areas in the Pan-Tropics and International Aid for Conservation." Geomatics 1, no. 3: 335-346.

Journal article
Published: 18 February 2021 in International Journal of Applied Earth Observation and Geoinformation
Reads 0
Downloads 0

Crop type mapping currently represents an important problem in remote sensing. Accurate information on the extent and types of crops derived from remote sensing can help managing and improving agriculture especially for developing countries where such information is scarce. In this paper, high-resolution RGB drone images are the input data for the classification performed using a transfer learning (TL) approach. VGG16 and GoogLeNet, which are pre-trained convolutional neural networks (CNNs) used for classification tasks coming from computer vision, are considered for the mapping of the crop types. Thanks to the transferred knowledge, the proposed models can successfully classify the studied crop types with high overall accuracy for two considered cases, achieving up to almost 83% for the Malawi dataset and up to 90% for the Mozambique dataset. Notably, these results are comparable to the ones achieved by the same deep CNN architectures in many computer vision tasks. With regard to drone data analysis, application of deep CNN is very limited so far due to high requirements on the number of samples needed to train such complicated architectures. Our results demonstrate that the transfer learning is an efficient way to overcome this problem and take full advantage of the benefits of deep CNN architectures for drone-based crop type mapping. Moreover, based on experiments with different TL approaches we show that the number of frozen layers is an important parameter of TL and a fine-tuning of all the CNN weights results in significantly better performance than the approaches that apply fine-tuning only on some numbers of last layers.

ACS Style

Artur Nowakowski; John Mrziglod; Dario Spiller; Rogerio Bonifacio; Irene Ferrari; Pierre Philippe Mathieu; Manuel Garcia-Herranz; Do-Hyung Kim. Crop type mapping by using transfer learning. International Journal of Applied Earth Observation and Geoinformation 2021, 98, 102313 .

AMA Style

Artur Nowakowski, John Mrziglod, Dario Spiller, Rogerio Bonifacio, Irene Ferrari, Pierre Philippe Mathieu, Manuel Garcia-Herranz, Do-Hyung Kim. Crop type mapping by using transfer learning. International Journal of Applied Earth Observation and Geoinformation. 2021; 98 ():102313.

Chicago/Turabian Style

Artur Nowakowski; John Mrziglod; Dario Spiller; Rogerio Bonifacio; Irene Ferrari; Pierre Philippe Mathieu; Manuel Garcia-Herranz; Do-Hyung Kim. 2021. "Crop type mapping by using transfer learning." International Journal of Applied Earth Observation and Geoinformation 98, no. : 102313.

Technical note
Published: 18 January 2021 in Remote Sensing
Reads 0
Downloads 0

The importance of tourism for development is widely recognized. Travel restrictions imposed to contain the spread of COVID-19 have brought tourism to a halt. Tourism is one of the key sectors driving change in Africa and is based exclusively on natural assets, with wildlife being the main attraction. Economic activities, therefore, are clustered around conservation and protected areas. We used night-time light data as a proxy measure for economic activity to assess change due to the pandemic. Our analysis shows that overall, 75 percent of the 8427 protected areas saw a decrease in light intensity in varying degrees in all countries and across IUCN protected area categories, including in popular protected area destinations, indicating a reduction in tourism-related economic activities. As countries discuss COVID-19 recovery, the methods using spatially explicit data illustrated in this paper can assess the extent of change, inform decision-making, and prioritize recovery efforts.

ACS Style

Anupam Anand; Do-Hyung Kim. Pandemic Induced Changes in Economic Activity around African Protected Areas Captured through Night-Time Light Data. Remote Sensing 2021, 13, 314 .

AMA Style

Anupam Anand, Do-Hyung Kim. Pandemic Induced Changes in Economic Activity around African Protected Areas Captured through Night-Time Light Data. Remote Sensing. 2021; 13 (2):314.

Chicago/Turabian Style

Anupam Anand; Do-Hyung Kim. 2021. "Pandemic Induced Changes in Economic Activity around African Protected Areas Captured through Night-Time Light Data." Remote Sensing 13, no. 2: 314.

Preprint
Published: 21 December 2020
Reads 0
Downloads 0

Evaluation of the effectiveness of protected areas is critical for forest conservation policies and priorities. To evaluate their effectiveness, we used 30-m resolution forest cover change data between 1990 and 2010 for ~4,000 protected areas and analyzed the relationships of the effectiveness of protected areas with socio-economic variables. Our results show that protected areas in the Tropics avoided 83,500 ± 21,200 km2 of deforestation during the 2000s. Brazil’s protected areas have the largest amount of avoided deforestation of 50,000 km2. We also show the amount of international aid received by tropical countries compared to the effectiveness of protected areas. International aid had major benefits in Latin America led by Brazil while tropical Asian countries used the resource ineffectively. Our results demonstrate that protected areas have been relatively more efficient in countries where deforestation pressures were increasing, and governance and forest change monitoring capacity are important factors for enhancing the efficacy of international aid.

ACS Style

Do-Hyung Kim; Anupam Anand. Effectiveness of Protected Areas in the Pan-Tropics and International Aid for Conservation. 2020, 1 .

AMA Style

Do-Hyung Kim, Anupam Anand. Effectiveness of Protected Areas in the Pan-Tropics and International Aid for Conservation. . 2020; ():1.

Chicago/Turabian Style

Do-Hyung Kim; Anupam Anand. 2020. "Effectiveness of Protected Areas in the Pan-Tropics and International Aid for Conservation." , no. : 1.