Convolutional Fuzzy Machine Learning based Algorithms for Mapping Chickpea Agricultural Crop Fields using Temporal Satellite Data

Abstract

In this research, we employed the convolutional classifiers, namely Fuzzy Local Information c-Means (FLICM) and Adaptive Fuzzy Local Information c-Means (ADFLICM), to classify and map chickpea agricultural fields in the Nagaur region of Rajasthan. This classification was performed using Planet Scope and Sentinel-2 data. To streamline the processing of temporal data and reduce spectral information, we experimented with the Normalized Difference Vegetation Index (NDVI) indices. To optimize the performance of the FLICM and ADFLICM algorithms, we utilized the Mean Membership Difference (MMD) technique to determine the most suitable fuzziness value. We also estimated the area of the ground-truthed chickpea fields to compare the performance of both methods. For the FLICM and ADFLICM algorithms with Planet Scope data, the estimated areas were 8914 m2 and 11278 m2, respectively. The ADFLICM algorithm was also applied to classify Sentinel-2 data, demonstrating its effectiveness with medium-resolution data, and the estimated area for Sentinel-2 data was 10633 m2. Finally, we calculated the Root Mean Square Error (RMSE) for the ADFLICM algorithm, which was found to be 0.075194, using Planet Scope satellite data as the reference dataset and Sentinel-2 satellite data as the classified dataset.