.png){kind=icon}
.png){kind=icon}
Return
A Lightweight Multiscale-Multiobject Deep Segmentation Architecture for UAV-Based Consumer Applications
.png){kind=icon}
PRE
AI
DOI：10.1109/TCE.2024.3367531.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Smart UAVs have been developed under the consumer Internet of Drone Things (CIoDTs) framework to improve the quality of service (QoS) for several commercial and consumer applications. Artificial intelligence (AI)-inspired algorithms are employed here to make these remote sensing devices more intelligent and agile to perform the task most effectively. However, these AI-based techniques may suffer from obtaining the required feature space, leading to poor performance of the AI system. Thus, to address these intrinsic issues, this manuscript presents a lightweight multiscale convolutional neural network (MCNN), which segments the UAV-captured aerial images based on their object classes. The proposed framework is based on an encoder-decoder architecture developed using depthwise separable convolution. The encoder module here involves extracting the crucial scale-invariant features that are upsampled at the decoder stage to generate the predicted segmented images. Additionally, the proposed architecture is validated using two datasets: the prepared NITRDrone dataset and the Urban Drone Dataset (UDD). Thus, the proposed method achieves an IoU score of 0.877 and 0.802 with the NITRDrone and UDD, respectively. Thus, the proposed lightweight architecture helps enhance the efficiency of these intelligent mobile edge aerial devices for several critical consumer applications demanding real-time data processing.
Keywords:
Task analysis
Remote sensing
Computer architecture
Autonomous aerial vehicles
Convolutional neural networks
Consumer electronics
Computer vision
VHR
remote sensing
aerial image
multiscale CNN
semantic segmentation
mobile edge devices
Journal
IF:
10.9
Papers:
5.1K
Citations:
6.8K
View more organizations