Evaluating green cover and open spaces in informal settlements of Mumbai using deep learning

Dabra, Ayush; Kumar, Vaibhav

doi:10.1007/s00521-023-08320-7

“…The evaluation metrics in Table 3 and Table 5 demonstrate the effectiveness of the proposed model in accurately generating the realistic images that can be used for accurate segmentation task. The overall performance displayed in Table 2 and Table 3 is higher than the overall performance achieved by all three segmentation models utilized in Dabra et al (2023) using the Greater Mumbai region dataset [47,48]. This verifies the use of synthetically generated data to improve the ability of deep learning models to perform well on unobserved and unseen data, resulting in better performance.…”

Section: Segmentation Using Real and Generated Satellite Imagesmentioning

confidence: 70%

“…We also discuss the potential impact of generating such realistic satellite images in the field of urban planning and growth, allowing informed decisions to be made. Furthermore, we compare the generated dataset of satellite images from both pix2pix and MD-pix2pix for the segmentation task and evaluate their performance against the benchmark set by Dabra et al ( 2023) [48] for real satellite images. By doing so, we aim to determine the quality of the generated dataset and establish its reliability for use in various applications.…”

Section: Resultsmentioning

confidence: 99%

“…The patches are extracted with a stride of 256 to have zero vertical and horizontal overlapping within the images from the 110 tiles of 600 × 600 dimension. The number of manually labeled patches is further increased by finetuning the predictions of the VGG16-UNet model used in Dabra et al (2023) [48], resulting in a total of 22,397 annotated patches. These corresponding paired patches are then subjected to affine transformations, horizontal and vertical flipping along with random cropping.…”

Section: Training Data Preparation and Preprocessingmentioning

confidence: 99%

“…The selection of models is done on the basis of previously available benchmarks on the original dataset used in Dabra et al (2023) [47,48]. The dataset for validating the use of generated satellite images in the segmentation task is made using 9378 pairs of satellite images and corresponding semantic segmentation.…”

Section: Segmentation Using Real and Generated Satellite Imagesmentioning

confidence: 99%

See 2 more Smart Citations

Generation Of Dense Urban Features Using Conditional GAN

Nihal,

Kumar

2023

Preprint

Self Cite

0

View full text Add to dashboard Cite

This paper discusses the use of conditional Generative Adversarial Networks (GANs) to generate dense urban features in satellite images and evaluate their effectiveness in semantic segmentation tasks. High-resolution true-color satellite imagery of Mumbai, obtained from Pleiades-1A at a 0.5m resolution, is utilized for the study. The proposed Multiple Discriminator pix2pix (MD-pix2pix) model, which employs multiple discriminators and a modified training procedure, is introduced to generate realistic satellite images. The performance of the MD-pix2pix model is compared to the traditional pix2pix model using a mix dataset of real and generated satellite images. The synthesized satellite images are assessed for their effectiveness in semantic segmentation tasks using various CNN models, including VGG16-UNet, MobileNetV2-UNet, and DeepLabV3+. The study aims to overcome the limitations of existing datasets that do not include informal settlements, such as slum areas, which are common in many cities. The results indicate that the MD-pix2pix model produces more realistic satellite images with greater variability in vegetation type, slum arrangements, and built-ups than the traditional pix2pix model. The synthetically generated satellite images are also effective in semantic segmentation tasks, with better segmentation accuracy achieved using the MD-pix2pix generated images for computationally less expensive architectures such as MobileNetV2-UNet. This study highlights the potential of GANs to generate realistic satellite images for urban feature mapping and monitoring applications.

show abstract

“…We also discuss the potential impact of generating such realistic satellite images in the field of urban planning and growth, allowing informed decisions to be made. Furthermore, we compare the generated dataset of satellite images from both pix2pix and MD-pix2pix for the segmentation task and evaluate their performance against the benchmark set by Dabra and Kumar (2023a) for real satellite images. By doing so, we aim to determine the quality of the generated dataset and establish its reliability for use in various applications.…”

Section: Resultsmentioning

confidence: 99%

Generation Of Dense Urban Features Using Conditional GAN

Nihal,

Kumar

2023

Preprint

Self Cite

0

View full text Add to dashboard Cite

This paper discusses the use of conditional Generative Adversarial Networks (GANs) to generate dense urban features in satellite images and evaluate their effectiveness in semantic segmentation tasks. High-resolution true-color satellite imagery of Mumbai, obtained from Pleiades-1A at a 0.5m resolution, is utilized for the study. The proposed Multiple Discriminator pix2pix (MD-pix2pix) model, which employs multiple discriminators and a modified training procedure, is introduced to generate realistic satellite images. The performance of the MD-pix2pix model is compared to the traditional pix2pix model using a mix dataset of real and generated satellite images. The synthesized satellite images are assessed for their effectiveness in semantic segmentation tasks using various CNN models, including VGG16-UNet, MobileNetV2-UNet, and DeepLabV3+. The study aims to overcome the limitations of existing datasets that do not include informal settlements, such as slum areas, which are common in many cities. The results indicate that the MD-pix2pix model produces more realistic satellite images with greater variability in vegetation type, slum arrangements, and built-ups than the traditional pix2pix model. The synthetically generated satellite images are also effective in semantic segmentation tasks, with better segmentation accuracy achieved using the MD-pix2pix generated images for computationally less expensive architectures such as MobileNetV2-UNet. This study highlights the potential of GANs to generate realistic satellite images for urban feature mapping and monitoring applications.

show abstract

Combining Local Knowledge with Object-Based Machine Learning Techniques for Extracting Informal Settlements from Very High-Resolution Satellite Data

Alrasheedi,

Dewan,

El-Mowafy

2024

Earth Syst Environ

0

View full text Add to dashboard Cite

The classification of informal settlements using very high-resolution (VHR) satellite data and expert knowledge has proven very useful for urban planning. The objective of this work was to improve the accuracy of informal settlement classification within the city of Riyadh, Saudi Arabia. The analysis incorporated the use of expert knowledge (EK). Twenty unique indicators relevant to informal settlements were identified by experts familiar with these areas, and incorporated into the image classification process. Object-based image analysis (OBIA) was then used to extract informal settlement indicators from a VHR image. These indicators were used to classify the image utilising two machine learning (ML) algorithms, random forest (RF) and support vector machine (SVM) methods. A VHR image (e.g., Worldview 3) of the city was employed. A total of 6,000 sample points were randomly generated, with 1800 used for training the VHR image. The classification process was able to clearly distinguish the formal settlement areas from informal areas, road networks, vacant blocks, shaded areas, and vegetation features. The object-based RF technique provided an overall accuracy of 96% (kappa value of 95%), while OB-SVM provided an accuracy of 95% (kappa of 91%). The results demonstrated that object-based ML methods such as RF and SVM, when combined with EK, can effectively and efficiently distinguish informal settlements from other urban features. This technique has the potential to be very useful for mapping informal settlements.

show abstract

Evaluating green cover and open spaces in informal settlements of Mumbai using deep learning

Cited by 7 publications

References 46 publications

Generation Of Dense Urban Features Using Conditional GAN

Generation Of Dense Urban Features Using Conditional GAN

Generation Of Dense Urban Features Using Conditional GAN

Combining Local Knowledge with Object-Based Machine Learning Techniques for Extracting Informal Settlements from Very High-Resolution Satellite Data

Contact Info

Product

Resources

About