Back to Search
Start Over
Subsurface drainage pipe detection using an ensemble learning approach and aerial images.
- Source :
-
Agricultural Water Management . Sep2023, Vol. 287, pN.PAG-N.PAG. 1p. - Publication Year :
- 2023
-
Abstract
- Subsurface drainage pipes are commonly used in the Midwestern United States to reduce excess soil moisture and improve crop yields. However, they are the considerable source of nonpoint pollution due to nutrient losses. Detecting the locations of drainage pipes is crucial for water quality management, but information about drainage pipe maps is often privately owned and unavailable. In this study, we propose an ensemble learning approach that uses eight fully convolutional networks (FCNs), including well-known architectures such as Unet, DenseNet, and Wnet, to detect subsurface drainage pipe locations from aerial images. Each FCN model is trained and validated using an aerial image dataset, taking a 256 × 256 × 3 pixel aerial image patch as input and outputting a pixel-wise drainage pipe detection map. Weighted averaging is then applied to the individual FCN outputs to create a unified drain pipe detection map. The performance of the proposed approach is evaluated using large-scale aerial image data that has not been used during the training and validation phases. The results demonstrate that the proposed approach provides accurate and more robust drain pipe detection over the case of using an individual FCN model. We further explore the effects of image resolution for the effective use of the proposed drain pipe detection approach. • Our ensemble learning method detects subsurface pipes accurately and robustly. • The ensemble model outperforms image processing and single machine learning models. • An optimal resolution can enhance the performance of the ensemble learning model. [ABSTRACT FROM AUTHOR]
Details
- Language :
- English
- ISSN :
- 03783774
- Volume :
- 287
- Database :
- Academic Search Index
- Journal :
- Agricultural Water Management
- Publication Type :
- Academic Journal
- Accession number :
- 169831110
- Full Text :
- https://doi.org/10.1016/j.agwat.2023.108455