Exploring the use of satellite imagery and computer vision‐based machine learning method to improve the spatial granularity of poverty statistics