2018.09.25 22:18 from 분류없음

플리커 사진으로 생태계 서비스 매핑하고 연결망 분석한 논문 나왔습니다. 

요새 한 반 년은 뭘 했나 돌아보면, CNN이랑 DQN 한다고 모 많이 하고 있고 (텐서플로우로 구현하고 우리 기존에 가지고 있는 생태계 모델들이랑 엮는 작업), NVIDIA GPU Grant 신청해서 받았고, 원격탐사 학회로 스페인 한 번 다녀왔고 어.. EGU 올해 다녀왔고, 다시 사회과학 공부를 좀 해야 해서 제도 경제학 책 읽고 그러는 중입니다. 이탄습지 연구도 참여해서 필드했고, 여러가지 여튼 최대한 ML/AI를 써서 생태계 연구, 토지 이용 연구에 붙여 보려고 발버둥.. 계약 기간 3년 중에 벌써 1년이 갔네요 그러다 보니. 



Mapping cultural ecosystem services 2.0 – Potential and shortcomings from unlabeled crowd sourced images 


We introduce an approach to a content analysis of geotagged photos for CES uses.

By using automated tags and a network analysis, themes of the photos were grouped.

This method allowed to distinguish CES- and non-CES-related photos.

This approach can provide spatial information about socio-cultural uses.

Our approach is applicable for crowd-sourced photos available in other regions.


The volume of accessible geotagged crowdsourced photos has increased. Such data include spatial, temporal, and thematic information on recreation and outdoor activities, thus can be used to quantify the demand for cultural ecosystem services (CES). So far photo content has been analyzed based on user-labeled tags or the manual labeling of photos. Both approaches are challenged with respect to consistency and cost-efficiency, especially for large-scale studies with an enormous volume of photos. In this study, we aim at developing a new method to analyze the content of large volumes of photos and to derive indicators of socio-cultural usage of landscapes. The method uses machine-learning and network analysis to identify clusters of photo content that can be used as an indicator of cultural services provided by landscapes. The approach was applied in the Mulde river basin in Saxony, Germany. All public Flickr photos (n = 12,635) belonging to the basin were tagged by deep convolutional neural networks through a cloud computing platform, Clarifai. The machine-predicted tags were analyzed by a network analysis that leads to nine hierarchical clusters. Those clusters were used to distinguish between photos related to CES (65%) and not related to CES (35%). Among the nine clusters, two clusters were related to CES: ‘landscape aesthetics’ and ‘existence’. This step allowed mapping of different aspects of CES and separation of non-relevant photos from further analysis. We further analyzed the impact of protected areas on the spatial pattern of CES and not-related CES photos. The presence of protected areas had a significant positive impact on the areas with both ‘landscape aesthetics’ and ‘existence’ photos: the total number of days in each mapping unit where at least one photo was taken by a user (‘photo-user-day’) increased with the share of protected areas around the location. The presented approach has shown its potential for reliable mapping of socio-cultural uses of landscapes. It is expected to scale well with large numbers of photos and to be easily transferable to different regions.

Posted by 관리자.. 트랙백 0 : 댓글 0


2018.02.02 00:07 from 분류없음

큰 일이 많았다. 몇 년에 한 번씩 오는 격동의 해. 

춘천일 마치고 여기로 왔다. 뮌헨 근처고, 연구소. 좋은 곳이고 동료들도 다들 뛰어나다. 분발하게 만드는 곳. 큰 이사를 세 번이나 했다 2017년엔. 짐 싸고, 풀고. 와선 다행히 쉽게 적응한 편이고, 언어도 매주 두 번씩 연구소에서 하니까 빨리 늘고 있다. 커스틴 선생님께 올초에 배우고 이번에 학원은 두 번째. 처음 독일 왔을 때 어학원을 제대로 다녔어야 했는데.. 죽기 전에 그래도 마스터 할 수 있을 것 같아 다행이다. 독어 마치면 스키를 좀 배워 보려고 그리고. 

친구가 죽었다. 가까웠던 친구고, 비슷하게 포닥을 하고 있었어서, 그리고 같이한 기억이 많아 충격량이 컸다. 올해의 큰 일 중 하나가 정식 가톨릭 신자가 된 거였는데, 무척 다행이고, 힘이 된다. 동네 성당에 매주 나가 기도하고 있다. 비보를 접하고 모든게 순식간에 낯설어졌는데, 동네 성당에서 부터 다시 실마리를 찾아 돌아 들어가고 있다. 당연한 얘기지만, 낯설게 보이는 사람은 나 혼자 뿐이지만, 괜찮았다 그동안. 성당이니까. 

한참 놓고 살다 요 근래 책을 좀 잡아서 JJ선생님과 홈선생 책을 거의 다 마쳤다. 에코 책을 다시 기차에서 읽기 시작했고 그리고, 돌아가려고 하고 있다 기억나는 그 날로, 책을 컴퓨터 보다 더 자주 찾던 날로. 아내가 졸업해서 본엘 다녀왔고, 잘 마쳤고 그리고. 논문 몇 개 리뷰받고 고치고, 나가기도 하고 그랬다. 여기서 일 하면서 새로 공부도 많이 하고 그래. 

괜찮아, 잘 할거야. 

Posted by 관리자.. 트랙백 0 : 댓글 0


2018.02.01 23:56 from 분류없음

Classification of rare land cover types: Distinguishing annual and perennial crops in an agricultural catchment in South Korea 

드디어 플로스원 논문 나왔습니다 



Many environmental data are inherently imbalanced, with some majority land use and land cover types dominating over rare ones. In cultivated ecosystems minority classes are often the target as they might indicate a beginning land use change. Most standard classifiers perform best on a balanced distribution of classes, and fail to detect minority classes. We used the synthetic minority oversampling technique (smote) with Random Forest to classify land cover classes in a small agricultural catchment in South Korea using modis time series. This area faces a major soil erosion problem and policy measures encourage farmers to replace annual by perennial crops to mitigate this issue. Our major goal was therefore to improve the classification performance on annual and perennial crops. We compared four different classification scenarios on original imbalanced and synthetically oversampled balanced data to quantify the effect of smote on classification performance. smote substantially increased the true positive rate of all oversampled minority classes. However, the performance on minor classes remained lower than on the majority class. We attribute this result to a class overlap already present in the original data set that is not resolved by smote. Our results show that resampling algorithms could help to derive more accurate land use and land cover maps from freely available data. These maps can be used to provide information on the distribution of land use classes in heterogeneous agricultural areas and could potentially benefit decision making.

Posted by 관리자.. 트랙백 0 : 댓글 0