Joint Multitask Learning for Image Segmentation and Salient Object Detection in Hyperspectral Imagery

With technological advancements, combining information from various tasks has become increasingly important. However, most feature learning approaches still focus on single-task learning. To address this, we propose a multitask learning-based model that simultaneously performs segmentation and salie...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Workshop on Hyperspectral Image and Signal Processing, Evolution in Remote Sensing S. 1 - 5
Hauptverfasser: Chhapariya, Koushikey, Ientilucci, Emmett J., Benoit, Alexandre, Buddhiraju, Krishna Mohan, Kumar, Anil
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 09.12.2024
Schlagworte:
ASPP
Attention Network
Computational modeling
Computer vision
Data models
Feature extraction
Hyperspectral Image Classification
Hyperspectral imaging
Image edge detection
Image segmentation
Multi-task Learning
Object detection
Optimization
Remote Sensing
Salient Object Detection
Shared-encoder decoder
Tuning
ISSN:2158-6276
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:With technological advancements, combining information from various tasks has become increasingly important. However, most feature learning approaches still focus on single-task learning. To address this, we propose a multitask learning-based model that simultaneously performs segmentation and saliency estimation. Our model is evaluated on two hyper-spectral datasets: HS-SOD for computer vision and Pavia University (PU) for remote sensing, which demonstrate strong generalization capabilities. By utilizing the additional spectral dimension in hyperspectral data, the model improves its ability to distinguish between materials and objects, leading to higher accuracy. The architecture features a shared encoder-decoder structure for efficiency, with an attention block enhancing segmentation by capturing key spectral-spatial features and a dense ASPP block improving salient object detection through multi-scale context. Extensive testing shows our model out-performs single-task approaches and state-of-the-art methods, proving its effectiveness and efficiency.
ISSN:2158-6276
DOI:10.1109/WHISPERS65427.2024.10876466