Document Type : Research Paper
Authors
1
Nuclear Agriculture Research School, Nuclear Science and Technology Research Institute (NSTRI), Karaj, Iran
2
Department of Irrigation & Reclamation Engineering, Faculty of Agriculture Engineering & Technology, University of Tehran, Karaj, Iran
3
Department of Agricultural Machinery Engineering, College of Agriculture and Natural Resources, University of Tehran, Karaj, Iran
4
Signal Processing & Machine Learning Section, Department of Electrical and Computer Engineering, Aarhus University, Aarhus, Denmark
Abstract
This study investigates the effectiveness of combining image processing techniques and machine learning methods to estimate the Manning roughness coefficient in furrow irrigation during the advance and storage phases. For this purpose, various input discharge values, irrigation cycles, phases, and soil texture types were considered. Images of the furrow surface were captured before and after each irrigation event, and the roughness coefficient in the advance and storage phases was estimated using the SIPAR_ID model and the Manning equation, respectively. Based on this data, an algorithm was developed that integrated image processing techniques with machine learning methods and was tested in three different scenarios.
The results showed that the algorithm, when using either images or field data separately, could not be properly trained and had very low accuracy, as some features were only accessible from images and others from field data. The results also revealed that the algorithm, when combining images with certain field data such as flow cross-section and discharge, performed very well in estimating the Manning roughness coefficient during both the advance and storage phases. In this scenario, the Random Forest and CART methods, with precision, recall, and F1-score values of 95%, 96%, and 95% respectively, outperformed other machine learning methods in estimating the Manning roughness coefficient.
Finally, it was suggested that similar studies be conducted considering other factors affecting roughness under different conditions, and that the algorithm be retrained accordingly to improve its performance and comprehensiveness.
Keywords
Main Subjects
)