EoCD: Encoder only Remote Sensing Change Detection

TL;DR

EoCD tackles the computational burden of remote sensing CD by adopting an encoder-only, early-fusion approach that omits a learnable decoder. A non-parametric EMFF module preserves multiscale semantic information, while a teacher–student distillation framework transfers guidance from a Siamese-encoder teacher to a decoder-free student. Across four public CD datasets and multiple encoders, EoCD achieves competitive or superior accuracy with substantially reduced complexity, highlighting that CD performance is largely driven by the encoder. The work delivers a practical pathway to faster, scalable CD in real-world remote sensing applications while maintaining high detection quality.

Abstract

Being a cornerstone of temporal analysis, change detection has been playing a pivotal role in modern earth observation. Existing change detection methods rely on the Siamese encoder to individually extract temporal features followed by temporal fusion. Subsequently, these methods design sophisticated decoders to improve the change detection performance without taking into consideration the complexity of the model. These aforementioned issues intensify the overall computational cost as well as the network's complexity which is undesirable. Alternatively, few methods utilize the early fusion scheme to combine the temporal images. These methods prevent the extra overhead of Siamese encoder, however, they also rely on sophisticated decoders for better performance. In addition, these methods demonstrate inferior performance as compared to late fusion based methods. To bridge these gaps, we introduce encoder only change detection (EoCD) that is a simple and effective method for the change detection task. The proposed method performs the early fusion of the temporal data and replaces the decoder with a parameter-free multiscale feature fusion module thereby significantly reducing the overall complexity of the model. EoCD demonstrate the optimal balance between the change detection performance and the prediction speed across a variety of encoder architectures. Additionally, EoCD demonstrate that the performance of the model is predominantly dependent on the encoder network, making the decoder an additional component. Extensive experimentation on four challenging change detection datasets reveals the effectiveness of the proposed method.

Figures (4)

• Figure 1: Comparison of various CD frameworks. a) In late fusion, $I_{pre}$ and $I_{post}$ are fed to Siamese encoder causing the backbone network to process each image separately which lead to increased computational cost. b) Early fusion approach prevent this by concatenating the bitemporal images before passing it to backbone, however, the sophisticated decoder still adds undesirable complexity to the model. c) EoCD introduces a simple design that bypasses the extra overhead of Siamese encoder and sophisticated decoder.
• Figure 2: Overall architecture of the proposed EoCD. It characterizes a student-teacher framework with a decoder-less student network. The student network performs early fusion of temporal images and optimally combines the multiscale representations, thereby significantly improving the efficiency of the network.
• Figure 3: Architecture of efficient multiscale feature fusion (EMFF) module. EMFF effectively combines the multiscale features while reducing the computational cost.
• Figure 4: Qualitative comparison of EoCD with BIT bit2022, ChangeFormer bandara2022changeformer, and ELGC-Net 2024_elgcnet CD methods. Data samples shown from row one to four correspond to LEVIR-CD, CDD-CD, SYSU-CD, and WHU-CD datasets, respectively. Notably, our approach demonstrates its capabilities to better detect the semantic changes highlighted in the yellow dotted boxes compared to existing methods.