Procreation of training data using cognitive science in temporal data processing for burnt paddy fields mapping

Abstract

While mapping specific classes such as burnt paddy fields at an interval require temporal remote sensing images that need ground truth data collection to be conducted during image acquisition. Frequent collection of such training data as per the frequency of acquisition of temporal data is time and cost dependent as well as cumbersome. The cognitive science concept has been applied to extend training data from seed training samples in the time domain to reduce ground visits for collecting training data. Since cognitive science is an interdisciplinary study of the mind and its processes, this research has tried to generate training samples in the past and the future within the paddy stubble burning cycle from the seed training samples collected from the ground. In this research work, seed training data collected on 25th October 2019 extended for 27th October 2019, 4th November 2019, 6th November 2019, as well as 20th October 2019, 17th October 2019, 15th October 2019 and 12th October 2019, temporal images for burnt field's identification. Favorable results were achieved through cognitive science using augmented training data in burnt paddy field mapping. The Mean Membership Difference between the training and testing site shows up to 98% accuracy. The largest burnt area was found between 4th Nov 2019 to 6th Nov 2019.The F-Score, Kappa and overall accuracy were found to be 98% in identifying the burnt paddy fields.