Advanced Land Classification using U-Net for Satellite Images

The well-known U-Net approach that has been used
successfully for healthcare imagery categorization, had been
developed and taught using the DeepGlobe Land Cover
Identification database for proper categorization of a variety of
different land cover categories, including cities, land used for
agriculture, wilderness, woodlands, bodies of water, other empty
land. The U-Net's encoder-decoder architecture with paired by
pass connections enables both exact localisation as well as
classification of different covering courses even in very
complicated and varied environments. The accuracy of the
model was tested both physically and numerically and it was
confirmed that it is capable of creating thorough and precise
segments maps which are essential for environmental tracking,
city development, particularly the control of resources. The
results show that the modified U-Net model can be successfully
employed to differentiate between different land covers, which
provides high-resolution insights that can be invaluable to policy
making and scientific research work in this regard, although this
study does not escape the difficulties that come with the land
cover classification task, like the need to be in possession of
larger and more diverse datasets to improve the generalization of
the model and increase its robustness. Future work will be done
to add domain-specific knowledge to increase model predictions
accuracy and conduct advanced research such as multi-spectral
data integration and transfer learning to enhance performance
even further. Additionally, the development of easy-to-use
visualization tools will be important to better engage
stakeholders, which will then be able to better interpret and
apply the model outputs in real world situations. This study
underlines the potential of deep learning models such as U-Net in
transforming the analysis of satellite images and contributing to
sustainable land management practices.Remarkably, the model
achieved an accuracy of 98.7% which is higher than the existing
methods such as RF, SVM and ANN, Faster R-CNN showing its
effectiveness and precision and implemented using Python.
Keywords - Land Cover Classification, U-Net Model, Satellite
Imagery, Environmental Monitoring, Remote Sensing.