Artificial intelligence is a field in full development, from facial recognition to autonomous vehicles and referral systems for online shopping, passing by smart farming, these new technologies are invading our daily lives.Nowadays, agricultural applications require more and more computer vision technologies for continuous monitoring and analysis of crop health and yield. That is why machine learning has become one of the mechanisms that make farming more efficient by using high-precision algorithms. This article deals with the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI), which are the most widely used indices in precision agriculture. In this work, we adopt GPU-based heterogeneous architecture using parallel programming with the CUDA language. The algorithm is evaluated on several platforms: NVIDIA Jetson TX1, DELL-desktop, and XU4 board. It has been discovered that the execution time of the two NDVI and NDWI indices on the embedded TX1 card is more optimized and improved compared to the execution time on the XU4 card and the Desktop.