A Stochastic Approach for Automatic Collection of Precise Training Data for a Soft Machine Learning Algorithm Using Remote Sensing Images

The quality of training data in terms of precision is instrumental in delivering a good performance for a supervised or a semi-supervised machine learning algorithm. The role of training data becomes more crucial for applications based on remote sensing data due to its distinct characteristics and limited amount of training data. In this paper, we present an effective approach for collection of precise training data from remote sensing images applied for a soft machine learning (softML) algorithm. The objective function of a fuzzy c-means algorithm was used to implement a softML algorithm driven by different spectral similarity measures for image classification and identification of wheat crop from remote sensing images. The spectral similarity measures selected for driving the softML algorithm were stochastic, deterministic as well as hybrid in nature. The training data for the soft computing algorithm were collected using manual and the proposed automatic mode. The manual training data were obtained by field survey of the study area, and the automatic training data were derived using a stochastic region growing algorithm. The precision of the training data obtained from both the modes were assessed by computing the classification accuracy for multiple classes using a fuzzy error matrix and analysis of membership distribution plots of the target wheat crop. Results from the experiments suggest that the proposed approach for automatic training data collection from remote sensing images results in an increment of the overall accuracy of classification for the softML algorithm. Also, the analysis of membership frequency distribution plots of the target wheat crop suggests greater precision of the training data collected by the proposed automatic mode. The increment in the frequency of the greater membership values and reduction in lower values indicated more accurate identification of the wheat crop using automatic mode of training the softML algorithm.