1P. Hima Bindu,2Bhusarapu Amrutha,3Iddum Bhavya Sri,4Surakasula Durga Prasad

Abstract

Remote sensing change detection is a fundamental technique in Earth observation applications, including
environmental monitoring, disaster assessment, and urban development analysis. The primary objective is to
identify differences between two satellite images of the same geographic region captured at different time
intervals. Conventional deep learning approaches mainly focus on spatial differences between images and
often fail to capture subtle or gradual changes because they do not fully utilize feature channel information
and temporal dependencies. This study proposes a deep learning–based change detection framework called
LENet, which integrates a Channel–Spatial Difference Weighting (CSDW) module and a layer-exchange
decoding structure. The framework processes bi-temporal satellite images using deep encoders to extract
discriminative feature maps. The CSDW module computes difference information across both spatial and
channel dimensions, enabling effective identification of structural and spectral variations, while the layerexchange
decoding mechanism strengthens interaction between temporal features and captures correlations
between images from the same geographical area. Experimental evaluations conducted on benchmark
datasets, including CLCD, PX-CLCD, LEVIR-CD, and S2Looking, demonstrate that the proposed model
achieves a change detection accuracy of 97.8%, producing more precise pixel-level change maps compared to
conventional CNN and Siamese architectures. The proposed system effectively detects subtle, gradual, and
large-scale changes, making it suitable for practical applications such as deforestation monitoring, urban
expansion analysis, and disaster damage assessment.