Abstract:
With the increasing number of satellites, remote sensing data is becoming more af fordable. This gratuitously available satellite imagery may be used as source data to
map Cultivated and Non-cultivated areas & crops types in agricultural fields. For such
mappings, previous developments included the use of basic ML methods and DL In the
present research study, we examine the models and approaches for semantic segmenta tion for the objective of Land Cover and Crop Type Classification. The manner in which
we will present our research involves two stages. (1) We prepare baseline on Cropland
Data Layer (CDL) provided by USDA. The semantic segmentation models are used for
classifying Cultivated and Non-cultivated Land Cover and out of the cultivated, crop
types are classified. (2) State-of-the-art (SOTA) semantic segmentation models lack in
manipulating the temporal dimension of time series imagery and evident solution to
process multi-spectral bands available in the satellite imagery. We propose a method ology to overcome these shortcomings by selecting appropriate band combinations for
crop type classification and treating time series visual data as a single image. Seasonal
Crop types data is collected from National Agriculture Research Center (NARC) Islam abad. The experimental findings illustrate that in CDL dataset, Land cover is classified
with an highest accuracy of nearly 90% and crop types with an accuracy of 70%, imply ing that the cropping patterns identified in one geographical area can be transferred to
another area, ensuring that such segmentation models can be utilized for real-time map ping tools in agriculture business & applications. On the other hand, the experimental
results yield 85% accuracy for classifying various crop types based on the evaluation of
NARC dataset of Pakistan Crop types.
