The Impact of Preprocessing by Contrast Enhancement on Spatial-temporal Attention Neural Network: An Evaluation in Remote Sensing Change Detection

Abstract

Remote sensing offers considerable advantages in detecting and monitoring the physical features of an area. There are remarkable studies in the literature geared towards developing robust machine learning models to automate area change detection based on remote sensing images. However, to date there lacks a detailed investigation into the impact of image enhancement techniques on machine learning models for remote sensing change detection. Remote sensing data is particularly limited to sufficient quality to support area monitoring. This study, therefore, aims to examine how significantly image contrast enhancement, with a focus on histogram matching and median filter techniques, contribute to the remote sensing classification performance. We utilize spatial-temporal attention neural network as the deep neural network-based detector model and conduct experiments on two benchmark datasets. Precision, recall, and F1-score are reported to evaluate the classification performance of the detector model with and without contrast enhancement as the preprocessing step.