A Two-Tiered Bidirectional Atrous Spatial Pyramid Pooling-Based Semantic Segmentation Model for Landslide Classification Using Remote Sensing Images

Abstract

Effective landslide representation from great spatial resolution images is significant in numerous applications. Many research works and techniques have been advertised. Still, these methods are very challenging to relate in real time since they depend on remotely sensing landslides from a solitary sensor with an exact spatial resolution. Precisely identifying landslides over a vast region with intricate background entities is difficult. Machine Learning (ML) and Deep Learning (DL) have attained extraordinary performance in classifying images utilizing remotely sensed images from numerous platforms. Moreover, techniques built within DL architectures tend to implement encoder-decoder network structures, where constant convolutions effortlessly strain out numerous landslide features. This study develops a Bidirectional Atrous Spatial Pyramid Pooling-Based Semantic Segmentation and Classification Model (BASPP-SSCM) technique for landslide Remote Sensing Images. The main goal of the BASPP-SSCM technique is to segment and classify the landslide areas. In the preprocessing stage, the BASPP-SSCM model employs an adaptive Wiener filtering (AWF) technique to eliminate the noise. Next, for the semantic segmentation method, the BASPP-SSCM technique utilizes the DeepLabV3 method with the backbone of the ConvNeXtLarge model for determining the landslide region. Furthermore, the CapsNet model is utilized for the feature extraction process. Besides, the Rigdelet neural network (RNN) technique is employed for the landslide classification process. At last, the pelican optimization algorithm (POA) methodology is implemented to fine-tune the parameters involved in the RNN model. A wide range of investigations is performed to highlight the superiority of the BASPP-SSCM method using a benchmark dataset. The performance validation of the BASPP-SSCM method underscored a superior accuracy value of 98.23% of other existing approaches.

Keywords

• Semantic segmentation
• landslide remote sensing images
• pelican optimization algorithm
• feature extraction
• atrous spatial pyramid pooling