Development of Attention Gated Trans-R2Unet with Novel Activation Function for Change Detection in Remote Sensing Images
Abstract
Change detection is considered to be a significant process which analyzes the transition of the land cover at different time intervals using high-resolution Remote Sensing Images (RSI). The RSIs generally have a synoptic view, maximum temporal frequency, spectral resolution and digital computation. Hence, several deep learning models have been developed by diverse researchers, which utilize neural networks and transformer models to perform automatic detection to evaluate the changes. However, the conventional techniques performed by analyzing the pixel patch changes and deep features lead to an error accumulation issue due to the intermediate process. To resolve such existing problems, a novel and intelligent framework is constructed for detecting changes in satellite images. To achieve this, the raw RSIs are collected from the standard online data sources. Subsequently, the obtained images are subjected to the model of Attention Gated Transformer-Recurrent Residual Unet with Exponential Linear Unit Activation Function (AGTR2U-ELUAF) for detecting the changes present in the images. This network is performed by influencing the concept of an attention-gated mechanism with a new activation function to get a better outcome. Finally, the performance of the system is validated using diverse performance measures and differentiated from other baseline methodologies. Hence, the findings reveal that it effectively and appropriately identifies the changes.