Improving Road Surface Area Extraction via Semantic Segmentation with Conditional Generative Learning for Deep Inpainting Operations

Abstract

The road surface area extraction task is generally carried out via semantic segmentation over remotely-sensed imagery. However, this supervised learning task is often costly as it requires remote sensing images labelled at the pixel level, and the results are not always satisfactory (presence of discontinuities, overlooked connection points, or isolated road segments). On the other hand, unsupervised learning does not require labelled data and can be employed for post-processing the geometries of geospatial objects extracted via semantic segmentation. In this work, we implement a conditional Generative Adversarial Network to reconstruct road geometries via deep inpainting procedures on a new dataset containing unlabelled road samples from challenging areas present in official cartographic support from Spain. The goal is to improve the initial road representations obtained with semantic segmentation models via generative learning. The performance of the model was evaluated on unseen data by conducting a metrical comparison where a maximum Intersection over Union (IoU) score improvement of 1.3% was observed when compared to the initial semantic segmentation result. Next, we evaluated the appropriateness of applying unsupervised generative learning using a qualitative perceptual validation to identify the strengths and weaknesses of the proposed method in very complex scenarios and gain a better intuition of the model’s behaviour when performing large-scale post-processing with generative learning and deep inpainting procedures and observed important improvements in the generated data.

A Survey of Deep Learning Road Extraction Algorithms Using High-Resolution Remote Sensing Images

“…The discriminator in their model used PatchGAN. Cira et al [ 110 , 111 ] applied the Pix2pix model to post-process road extraction. They improved the integrity of road surface area extraction by contaminating labels and reconstructing them.…”

“…The discriminator in their model used PatchGAN. Cira et al [110,111] applied the Pix2pix model to post-process road extraction. They improved the integrity of road surface area extraction by contaminating labels and reconstructing them In addition, Abdollahi et al [7] proposed a deep learning approach using conditional gen erative adversarial networks (CGANs) for road segmentation in high-resolution aerial im agery.…”

Dataset containing orthoimages tagged with road information covering approximately 8650 km2 of the Spanish territory (SROADEX)

“…In our previous works [1] , [2] , [3] , [4] , having used datasets containing around 10,000 tiles of 256 × 256 pixels from representative areas for road recognition and extraction, we observed that one of the main drawbacks of the training processes was the lack of sufficient data, as inconsistent predictions were frequently present. In order to improve the efficiency of road extraction solutions, we drastically increased the size of the datasets used for training (up to more than 500,000 tiles) by labelling samples from new regions of the Spanish territory.…”

Deep Dual-Resolution Road Scene Segmentation Networks Based on Decoupled Dynamic Filter and Squeeze–Excitation Module

“…The performance of deep learning methods in semantic segmentation [ 6 ] has exceeded that of traditional methods. Most image semantic segmentation networks [ 7 , 8 ] based on deep learning methods are based on Fully Convolutional Networks (FCNs) [ 9 ] proposed by Long et al…”

“…The performance of deep learning methods in semantic segmentation [6] has exceeded that of traditional methods. Most image semantic segmentation networks [7,8] based on deep learning methods are based on Fully Convolutional Networks (FCNs) [9] proposed by Long et al According to different types of roads, road scenes include urban streets, highways, rural roads, etc. Depending on the weather, road scenes include rain, snow, sunny days, or heavy fog.…”