Machine learning based crop/farm type classification using polarimetric SAR images
Information on the spatial distribution of crops is of great importance for the sustainable management and development of agricultural practices and thus for national and global economic development. Obviously, Earth observation data provide an optimal opportunity for the production and updating of land use maps over agricultural lands. Among the various remote sensing tools, polarimetric SAR images are known to be the most optimal technique for crop identification from space. The images are independent of atmospheric conditions and provide optimal information for distinguishing different types of plants. This proposal aims to use polarimetric SAR images to develop an automatic machine learning method to classify land uses and, in particular, crop types in any desire areas.
The main goal of this project is to develop a machine learning approach for correct cropland mapping using polarimetric SAR images. The idea is to develop a supervised machine learning technique such as support vector regression or deep learning to characterize different farmland in polarimetric images. For this purpose, existing Radrasat-2 images and land use maps in the Netherlands will be used to build and train the model. In particular, the proposal investigates how to optimally build and train the model. This requires indeed the use of an optimal number of training samples, optimal information extracted from polarimetric data and an efficient definition of the cost function in the training process. Once the model is built, it can be used to predict crop species in any desired area of the world. Students are encouraged to extend their results to their area of interest using freely available polarimetric Sentinel-1 images.
Ahishali, Mete, et al. "Classification of polarimetric SAR images using compact convolutional neural networks." GIScience & Remote Sensing 58.1 (2021): 28-47. https://www.tandfonline.com/doi/epub/10.1080/15481603.2020.1853948?needAccess=true
Aghababaei, H., & Sahebi, M. R. (2016). Wishart Derived Distance Based Clustering of Polarimetric SAR Images Using Support Vector Machines. Journal of the Indian Society of Remote Sensing, 44(6), 1003-1010
Cheng, J.; Zhang, F.; Xiang, D.; Yin, Q.; Zhou, Y.; Wang, W. PolSAR Image Land Cover Classification Based on Hierarchical Capsule Network. Remote Sens. 2021, 13, 3132. https://doi.org/10.3390/rs13163132