Comparative Analysis of aerial image segmentation using U-net with customized encoder and decoder networks

A critical problem in computer vision is pixel-level object classification and delineation, which is investigated in this study, "Implementation and comparative analysis of semantic segmentation using hybridized deep learning models." While convolutional neural nets (CNNs) have demonstrate...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2024 First International Conference on Innovations in Communications, Electrical and Computer Engineering (ICICEC) S. 1 - 5
Hauptverfasser: Prithvi, P, Kumar, P Perin
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 24.10.2024
Schlagworte:
Accuracy
Complexity theory
Computational modeling
Convolutional neural networks
Decoding
Deep learning
Drone dataset
Drones
mIoU
Pixel Accuracy
Semantic segmentation
Technological innovation
U-Net
Visualization
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:A critical problem in computer vision is pixel-level object classification and delineation, which is investigated in this study, "Implementation and comparative analysis of semantic segmentation using hybridized deep learning models." While convolutional neural nets (CNNs) have demonstrated promise in semantic segmentation, complex visual scenarios continue to present challenges for CNNs. Our unique approach proposes combining CNNs with Deep neural networks or attention mechanisms to combine deep learning models, allowing for the utilization of complementary information and an improvement in segmentation accuracy. The U-Net is hybridized utilizing models like mobilenetv2, vgg19, resnet152, inceptionnetv2, and densenet169 and the encoder decoder networks employed using these models are also customized in directive to reduce the complexity of the model. The aerial semantic segmentation drone dataset from Kaggle is used to assess the performance of the model. For evaluation, common criteria like mean intersection over union (mIoU) and pixel accuracy are employed.
DOI:10.1109/ICICEC62498.2024.10808956