한국산림과학회지 (Journal of Korean Society of Forest Science)

  • 제110권2호
  • /
  • Pages.155-164
  • /
  • 2021
  • /
  • 2586-6613(pISSN)
  • /
  • 2586-6621(eISSN)
한국산림과학회 (Korean Society of Forest Science)
권호 표지
DOI QR코드

DOI QR Code

수목 동정을 위한 수피 분류 데이터셋 구축과 합성곱 신경망 기반 53개 수종의 동정 모델 개발

Construction of a Bark Dataset for Automatic Tree Identification and Developing a Convolutional Neural Network-based Tree Species Identification Model

  • 김태경 (서울대학교 농림생물자원학부) ;
  • 백규헌 (서울대학교 산림과학부) ;
  • 김현석 (서울대학교 농림생물자원학부)
  • Kim, Tae Kyung (Department of Agriculture, Forestry and Bioresources, Seoul National University) ;
  • Baek, Gyu Heon (Department of Forest Sciences, Seoul National University) ;
  • Kim, Hyun Seok (Department of Agriculture, Forestry and Bioresources, Seoul National University)
  • 투고 : 2021.03.04
  • 심사 : 2021.06.01
  • 발행 : 2021.06.30
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

자연환경에 대한 국민들의 관심 증가로 스마트폰과 같은 휴대용 기기를 이용한 수목 동정의 자동화에 대한 요구가 증가하고 있다. 최근 딥러닝 기술의 발전에 힘입어, 외국에서는 수목 인식 분야에의 적용이 활발하게 이루어지고 있다. 수목의 분류를 위해 꽃, 잎 등 다양한 형질들을 대상으로 연구가 진행되고 있지만, 접근성을 비롯한 여러 장점을 가진 수피의 경우 복잡도가 높고 자료가 부족하여 연구가 제한적이었다. 본 연구에서는 국내에서 흔히 관찰 가능한 수목 54종의 사진자료를 약 7,000 여장 수집 및 공개하였고, 이를 해외의 20 수종에 대한 BarkNet 1.0의 자료와 결합하여 학습에 충분한 수의 사진 수를 가지는 53종을 선정하고, 사진들을 7:3의 비율로 나누어 훈련과 평가에 활용하였다. 분류 모델의 경우, 딥러닝 기법의 일종인 합성곱 신경망을 활용하였는데, 가장 널리 쓰이는 VGGNet (Visual Geometry Group Network) 16층, 19층 모델 두 가지를 학습시키고 성능을 비교하였다. 또한 본 모형의 활용성 및 한계점을 확인하기 위하여 학습에 사용하지 않은 수종과 덩굴식물과 같은 방해 요소가 있는 사진들에 대한 모델의 정확도를 확인하였다. 학습 결과 VGG16과 VGG19는 각각 90.41%와 92.62%의 높은 정확도를 보였으며, 더 복잡도가 높은 모델인 VGG19가 조금 더 나은 성능을 보임을 확인하였다. 학습에 활용되지 않은 수목을 동정한 결과 80% 이상의 경우에서 같은 속 또는 같은 과에 속한 수종으로 예측하는 것으로 드러났다. 반면, 이끼, 만경식물, 옹이 등의 방해 요소가 존재할 경우 방해요소가 자치하는 비중에 따라 정확도가 떨어지는 것이 확인되어 실제 현장에서 이를 보완하기 위한 방법들을 제안하였다.

Many studies have been conducted on developing automatic plant identification algorithms using machine learning to various plant features, such as leaves and flowers. Unlike other plant characteristics, barks show only little change regardless of the season and are maintained for a long period. Nevertheless, barks show a complex shape with a large variation depending on the environment, and there are insufficient materials that can be utilized to train algorithms. Here, in addition to the previously published bark image dataset, BarkNet v.1.0, images of barks were collected, and a dataset consisting of 53 tree species that can be easily observed in Korea was presented. A convolutional neural network (CNN) was trained and tested on the dataset, and the factors that interfere with the model's performance were identified. For CNN architecture, VGG-16 and 19 were utilized. As a result, VGG-16 achieved 90.41% and VGG-19 achieved 92.62% accuracy. When tested on new tree images that do not exist in the original dataset but belong to the same genus or family, it was confirmed that more than 80% of cases were successfully identified as the same genus or family. Meanwhile, it was found that the model tended to misclassify when there were distracting features in the image, including leaves, mosses, and knots. In these cases, we propose that random cropping and classification by majority votes are valid for improving possible errors in training and inferences.

키워드

과제정보

본 연구는 산림청(한국임업진흥원) 산림과학기술 연구개발사업(2018113B10-2020-BB01)과 산림과학기술 연구개발사업(2020185D10-2122-AA02)의 지원에 의하여 이루어진 것입니다.

참고문헌

  1. Blaanco, L.J., Travieso, C. M., Quinteiro, J. M., Hernandez, P. V., Dutta, M. K. and Singh, A. 2016. A bark recognition algorithm for plant classification using a least square support vector machine. Ninth International conference on contemporary computing 2016: 1-5.
  2. Boudra, S., Yahiaoui, I. and Behloul, A. 2015. A comparison of multi-scale local binary pattern variants for bark image retrieval. Computer Science 9386: 764-775.
  3. Bressane, A., Roveda, J.A.F. and Martins, A.C.G. 2015. Statistical analysis of texture in trunk images for biometric identification of tree species. Environmental Monitoring and Assessment 187(4): 212. https://doi.org/10.1007/s10661-015-4400-2
  4. Brodrick, P.G., Davies, A.B. and Asner, G.P. 2019. Uncovering ecological patterns with convolutional neural networks. Trends in Ecology & Evolution 34(8): 734-745. https://doi.org/10.1016/j.tree.2019.03.006
  5. Carpentier, M., Giguere, P. and Gaudreault, J. 2018. Tree species identification from bark images using convolutional neural networks. In 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS): 1075-1081.
  6. Chi, Z., Houqiang, L. and Chao, W. 2003. Plant species recognition based on bark patterns using novel Gabor filter banks. Neural Networks and Signal Processing 2: 1035-1038.
  7. Choi, J.E. 2019. A Tree classification model using CNN Inception v3. (Dissertation). Seoul. Ewha Womans' University.
  8. Cimpoi, M., Maji, S. and Vedaldi, A. 2015. Deep filter banks for texture recognition and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 3828-3836.
  9. Fiel, S. and Sablatnig, R. 2011. Automated identification of tree species from images of the bark, leaves and needles. Proceedings of the 16th Computer Vision winter workshop. pp. 67-74.
  10. Hayat, K. 2018. Multimedia super-resolution via deep learning: A survey. Digital Signal Processing 81: 198-217. https://doi.org/10.1016/j.dsp.2018.07.005
  11. Huang, Z.-k., Huang, D.-S., Du, J.-X., Quan, Z.-h. and Gua, S.-B. 2006. Bark classification based on contourlet filter features. Intelligent Computing. pp. 1121-1126.
  12. Kim, M.K. 2019. Bark identification using a deep learning model. Journal of Korea Multimedia Society 22(10): 1133-1141.
  13. Krizhevsky, A., Sutskever, I. and Hinton, G.E. 2012. Imagenet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems 25: 1097-1105.
  14. Mata-Montero, E., Carranza-Rojas, J. 2016. Automated plant species identification: challenges and opportunities. IFIP World Information Technology Forum. pp. 26-36.
  15. Min, A. 2020. A Study on Transfer Learning for Image Classification of Convolutional Neural Network: Based on VGG16 Deep Convolutional Neural Network. (Dissertation). Gunpo-si. Hansei University.
  16. Mizoguchi, T., Ishii, A., Nakamura, H., Inoue, T. and Takamatsu, H. 2017. Lidar-based individual tree species classification using convolutional neural network. Proceedings of the Society of Photo-optical Instrumentation Engineers 10332: 1-7.
  17. Park, K.H. 2013. Characteristics on Tree Shapes and Bark Types of Landscape Trees Species in Korea. (Dissertation). Gyeongsan-si. Yeungnam University.
  18. Ratajczak, R., Bertrand, S., Crispim, C.J. and Tougne, L. 2019. Efficient bark recognition in the wild. Proceedlings of the International Conference on Computer Vision Theory and Application. pp. 240-248.
  19. Simonyan, K. and Zisserman, A. 2015. Very deep convolutional networks for large-scale image recognition. International Conference on Learning Representations 2015: 2-8.
  20. Svab, M. 2014. Computer-vision-based tree trunk recognition. (Dissertation). Republic of Slovenia. Faculty of Computer and Information Science, University of Ljubljana.
  21. Yoon, Y.C., Sang, J.H. and Park, S.M. 2018. Trends of plant image processing technology. 2018 Electronics and Telecommunications Trends 33(4): 54-60.