Unet 모델을 활용한 토석채취지 모니터링 정확도 평가
Evaluation of Monitoring Accuracy for Quarry Using Unet Models
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 연구는 Segmentation 기반의 딥러닝 모델을 활용하여 토석채취지 모니터링 방안 검토를 연구의 목적으로 하였다. 딥러닝 모델은 베이스 모델로 Unet을 활용하였으며, DeeplabV3+ 모델의 일부 구조를 융합한 수정된 Unet 모델(Modified Unet) 두가지를 활용하였다. 또한, 질감정보 활용 유무에 따른 분류정확도 검토를 위해 딥러닝 데이터세 트를 가시광선 파장만을 활용한 DatasetA와 Entropy, STD 정보를 함께 활용한 DatasetB로 구성하였다. 딥러닝 모델 의 학습은 학습 및 검증 정확도가 안정화된 1,000 Epoch까지 진행하였다. 딥러닝 모델을 활용한 토석채취지 분류지도 와 참값인 라벨이미지를 비교·검토하여 정확도를 평가한 결과, Modified Unet 모델은 Unet 모델에 비해 탐지 정확도 가 향상되었음을 확인하였다. 특히 침사지 및 인공 구조물과 같은 범주에서는 Entropy, Standard Deviation (STD) 등의 질감 정보를 도입하여 분류 정확도가 더욱 향상되었음을 알 수 있었다. 반면, 계단식 채취지역, 묘목장 등의 범주에 서는 F1-Score가 약 50% 이하로 분포하였다. 이를 통해 가시광선 파장만을 활용한 모델에서는 분류에 한계점이 있다는 것을 확인하였으며, 추가적인 학습 자료 구축과 함께, 근적외선, 식생 지수, 지형 정보 등 다양한 데이터를 활용하여 모 델을 보완하는 것이 필요하다고 판단된다.
Abstract
The aim of this study was to explore segmentation-based deep learning models for quarry monitoring. In this study, we adopted the Unet model as the foundational architecture and employed two distinct deep learning models for our analysis: the conventional Unet model and a modified version of the Unet model that integrates certain structural components from the DeeplabV3+ model. To investigate the impact of texture information on classification accuracy, we assembled two distinct datasets: Dataset A, utilizing only visible light wavelengths, and Dataset B, which included additional Entropy and Standard Deviation (STD) data. The training of the deep learning models was conducted for up to 1,000 epochs, at during which point both the training and validation accuracies were observed to have stabilized. After comparing and evaluating the accuracy of classification maps generated using deep learning models with the reference label images, we discovered that the modified Unet model showed improved detection accuracy compared to the conventional Unet model. In specific categories, we observed that incorporating texture information, such as Entropy and Standard Deviation (STD), further enhanced classification accuracy. Conversely, in certain categories, the F1-Score was distributed below approximately 50%. Through this observation, we have identified limitations in classification when exclusively utilizing models based on visible light wavelengths.
참고문헌 (21)
- Basu, S., "Deepsat: A learning framework for satellite imagery", Proc. of the 23rd SIGSPATIAL International Conference on Advances in Geographic Information Systems, pp.37~ (2015)
- "What Is the Difference between Test and Validation Datasets?" (2022)[URL]
- Chen, L. C., "Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs", IEEE transactions on pattern analysis and machine intelligence, 40(4), pp.834~848 (2017)
- Choi, B. K., "Analysis of land-cover change and fragmentation pattern using Landsat TM data", Landscape and Geography, 23(1), pp.75~85 (2013)
- Choi, Y. H., "The Status of Rehabilitation Region on Artifical Slope", Proceeding of Korean Society of Environmental Conference, 19(2), pp.234~237 (2009)
- Gärtner, P., "Object based change detection of Central Asian Tugai vegetation with very high spatial resolution satellite imagery", International Journal of Applied Earth Observation and Geoinformation, 31, pp.110~121 (2014)
- Géron, A., "Hands-on machine learning with scikit-learn and tensorflow: Concepts. Tools, and Techniques to build intelligent systems" (2017), O’Reilly Media
- Han, Z., "Comparing fully deep convolutional neural networks for land cover classification with high-spatial-resolution Gaofen-2 images", ISPRS International Journal of Geo-Information, 9(8), pp.478~ (2020)
- Hu, B., "Improving urban land cover classification with combined use of sentinel-2 and sentinel-1 imagery", ISPRS International Journal of Geo-Information, 10(8), pp.533~ (2021)
- Huang, M. H., "Artificial intelligence in service", Journal of Service Research, 21(2), pp.155~172 (2018)
- Im, J., "Enhancing binary change detection performance using a moving threshold window(MTW)approach", Photogrammetric Engineering & Remote Sensing, 75(8), pp.951~961 (2009)
- Kim, J. S., "Assessing the applicability of UAV for Sea Cliff Erosion-rate Monitoring", Landscape and Geography, 27(2), pp.1~13 (2017)
- LeCun, Y., "Generalization and network design strategies", Connectionism in perspective, 19(143-155), pp.18~ (1989)
- Liu, D., "Assessing object-based classification : advantages and limitations", Remote sensing letters, 1(4), pp.187~194 (2010)
- Na, H. S., "A Study on Spatial Characteristic of Damaged Forest Areas by Image Interpretation-Focused on DMZ region in Hwacheon and Cheorwon", Landscape and Geography, 25(3), pp.63~73 (2017)
- Park, J. H., "Analysis of the Restoration Actual Conditions on the Quarries", Journal of the Korean Society of Environmental Restoration Technology, 13(3), pp.128~142 (2010)
- Park, J. H., "Analysis of the Restoration Actual Conditions on the Quarries II", Journal of Korean Forest Society, 101(4), pp.558~566 (2012)
- Park, J. H., "Analysis of quarrying and restoration characteristics on quarry in Korea", Journal of Korean Society of Forest Science, 105(2), pp.223~230 (2016)
- Ronneberger, O., "U-net: Convolutional networks for biomedical image segmentation", Proc. Of International Conference on Medical Image Computing and Computer-Assisted Intervention, pp.234~241 (2015), Springer
- Rouhi, R., "Benign and malignant breast tumors classification based on region growing and CNN segmentation", Expert Systems with Applications, 42(3), pp.990~1002 (2015)
- Schöpfer, E., "Remote sensing of urban and suburban areas", pp.181~192 (2010)