Sravya Nedungatt

Building object extraction plays a crucial role in the monitoring and strategic planning of urban infrastructures. With the rapid advancement of deep learning (DL), segmentation tasks are becoming more effective, even though models trained on datasets from different regions may not perform well in Indian areas due to their complex, unstructured cityscapes. Therefore, this paper introduces a novel building object segmentation dataset, namely, NITK Computer Vision and Deep Learning (NCVDL) Cartosat-2S building dataset, which covers three diverse regions of India. Moreover, this paper presents a lightweight DL model, namely, the KAN Building Detection Network (KANBDNet), for extracting buildings from satellite images, aiming to reduce the number of parameters without compromising performance using Kolmogorov–Arnold Networks, and the training is guided by a boundary-focused mIoU metric. It is mainly structured using three modules: a split-wise feature extraction module to efficiently extract features by splitting them into two and then process separately, a dense multiscale feature extraction block to extract and densely connect features at various scales, and a skip layer which effectively combines the encoder stage and high level features to enhance the features passing through the decoder. The performance of the proposed KANBDNet and state-of-the-art (SOTA) DL-based building segmentation models is evaluated using the new NCVDL Cartosat-2S dataset and the publicly available GF-7 building dataset. The proposed KANBDNet outperforms all SOTA models, achieving ( F 1 -score, kappa coefficient, jaccard index, dice coefficient) of (87.75%, 75.50%, 78.87%, 80.69%) and (92.10%, 84.19%, 85.83%, 86.62%) on the NCVDL Cartosat-2S and GF-7 building datasets, respectively.

Synthetic Aperture Radar(SAR) images are extensively used for Earth observation because of their all-weather, day, and night imaging capabilities. However, speckle noise in SAR images significantly reduces their usability in a variety of applications. Deep learning models developed for SAR despeckling exhibit promising noise reduction capabilities. Bringing a balance between reducing graininess and preserving texture details is a challenging task. In addition, supervised training of a robust deep learning model requires noisy images that capture the SAR speckle dynamics and the corresponding speckle-free ground truth, which is generally not available. This study proposes the first hybrid CNN-Halo attention-based transformer model for SAR despeckling. CNN-based feature extraction modules provide multiscale and multidirectional and large-scale feature maps. A halo-attention transformer block is used in the skip connection. It aids in the better preservation of radiometric information in the despeckled SAR images. TransSARNet is trained in a supervised manner using a new synthetic SAR dataset, which is a combination of the Kylberg and UCMerced land-use datasets. This study also analyzed the effect of combining the Kylberg and UCMerced datasets on texture preservation in despeckled SAR images. The visual and qualitative metrics evaluated on Sentinel-1 Single Look Complex SAR data showed that the proposed TransSARNet approach outperformed the other models under consideration. TransSARNet achieves a harmonious balance between model complexity, despeckling ability, edge preservation, radiometric information preservation, and smoothing in homogeneous regions.

Semantic change detection in remote sensing imagery plays a pivotal role in urban planning, environmental monitoring, and disaster assessment applications. Existing deep learning-based methods, particularly those relying on triple-branch architectures, often struggle to accurately localize and predict changes in complex spatial environments characterized by diverse land-cover types. To overcome these limitations, this paper proposes a novel network called the Spatial Flow-based Semantic Change Detection Network. This network processes bi-temporal satellite images using a dual-encoder, triple-decoder architecture that progressively refines spatial features at each network stage, improving semantic change detection results. The Attention-Based Siamese Encoder, Cascaded Convolutional Attention Fusion Block, Cascaded Convolutional Attention Refinement Block and Differentiable Binarization layer helps in improving semantic change detection performance. Experimental results of proposed network on the SECOND dataset demonstrate that the proposed model significantly improves the ability to localize critical changes and distinguish between change and no-change regions. The proposed network achieves an overall accuracy of 86.32%, a mean Intersection over Union of 70.33%, a Separated Kappa of 21.21%, and an F1-score for semantic change detection of 66.01%, with a score of 35.94%. These results represent substantial improvements over previous state-of-the-art models, including a 0.26% increase in overall accuracy, a 2.21% increase in mean Intersection over Union, a 2.62% enhancement in Separated Kappa, and a 3.6% improvement in F1-score for semantic change detection compared to the best-performing models. Notably, the proposed network achieves these results with only 14.56 million parameters, making it more effective and efficient than its competitors, which utilize over 22 million parameters.

Deep learning (DL) algorithms are currently the most effective methods for change detection (CD) from high-resolution multi-spectral (MS) remote-sensing (RS) images. Because a variety of satellites are able to provide a lot of data, it is now easy to find out changes using efficient DL models. Current CD methods focus on simple structure and combining the features obtained by all the stages together rather than extracting multi-scale features from a single stage since it may lead to information loss and imbalance contribution of features at different stages. This in turn results in misclassification of small changed areas, poor edge and shape preservation of changed areas. This paper introduces an Enhanced RSCD network (ERSCDNet) for CD from bi-Temporal (BT) aerial and MS images. The proposed encoder-decoder based ERSCDNet model uses an attention based encoder and decoder blocks and a Modified New Spatial Pyramid Pooling block (MNSPP) at each stage of the decoder part which effectively utilize features at each encoder stages and prevent information loss. The Learning, Vision and Remote Sensing - CD (LEVIR-CD), Onera Satellite Change Detection (OSCD) and Sun Yat-Sen University - CD (SYSU-CD) datasets are used to evaluate the ERSCDNet model. The ERSCDNet gives better performance than all the models used in this paper for comparison. It gives a F1 Score (F1), a Kappa Coefficient (Ka), and a Jaccard Index (JI) of (0.9306, 0.9282, 0.8703), (0.8945, 0.8887, 0.8091) and (0.7581, 0.6876, 0.6103) on OSCD, LEVIR-CD and SYSU-CD datasets respectively.