This research focuses on optimizing parallel k-means clustering for mapping paddy fields in Java using time-series satellite imagery from MODIS. The study employs wavelet transforms to reduce noise in the data and compares the efficiency of CPU and GPU processing methods, highlighting that an ARM-based Raspberry Pi system offers significant speed and energy-saving benefits. The findings indicate that both CPU and GPU architectures can effectively classify vegetation patterns, with Raspberry Pi demonstrating the highest performance efficiency.