This research explores the performance of existing classification models with remote-sensing satellite images of Bangladesh's roads. The models were analyzed under normal conditions, in addition, tried to break down the images in various methods. The goal is to find out the type of data that facilitates these models to give the best results and develop monitoring system. It is also investigated, what other things in the images, besides the roads, affect the models' performance. Datasets were created specifically focused on road classification and segmentation and tried various deep-learning models for this purpose. This research helps identify the best data types for improving classification accuracy and highlights what elements in the images can impair the models' functionality. In the end, we componentized the trained models utilizing web technologies and created automation systems for road quality measurement.