Edge-ANN: Storage-Efficient Edge-Based Remote Sensing Feature Retrieval

TL;DR

Edge-ANN tackles storage-constrained on-device remote sensing feature retrieval by replacing dimension-dependent hyperplanes with implicit partitions defined by pairs of anchors. It introduces Binary Anchor Optimization, consisting of Phase 1 iterative anchor search and Phase 2 hyperplane position fine-tuning, to construct balanced partitions with fixed-anchor references and a provable local optimum. The approach achieves 30–40% reduction in secondary storage with only 3–5% recall loss across MillionAID, Hi-UCD, and GUN datasets, and demonstrates superior performance under tight primary storage budgets compared with traditional tree- and graph-based ANN methods, while providing a practical phase diagram to guide algorithm selection. The work significantly lowers the storage footprint of large-scale, real-time remote sensing feature retrieval on edge devices, and lays groundwork for extending to additional distance metrics beyond Euclidean. The proposed method decouples model size from feature dimension, enabling scalable, on-device retrieval without sacrificing core accuracy or efficiency under constrained resources.

Abstract

Meeting real-time constraints for high-performance Approximate Nearest Neighbor (ANN) search remains a critical challenge in remote sensing edge devices, which are essentially fusion systems like micro-satellites and UAVs, largely due to stringent limitations in primary (RAM) and secondary (disk) storage. To address this challenge, we propose Edge-ANN, an innovative ANN framework specifically engineered for storage efficiency. The core innovation of Edge-ANN lies in its departure from traditional tree-based methods that store high-dimensional hyperplanes. Instead, it leverages pairs of existing data items, termed "anchors," to implicitly define spatial partitions. To ensure these partitions are both balanced and effective, we have developed a novel Binary Anchor Optimization algorithm.This architectural shift eliminates the dimension-dependence of the space complexity. Rigorous experiments on three multi-source datasets, MillionAID, High-resolution Urban Complex Dataset, and GlobalUrbanNet Dataset, demonstrate that under simulated edge environments with dual storage constraints, Edge-ANN achieves a 30-40% reduction in secondary storage compared to the baseline, at the cost of a minor 3-5% drop in retrieval accuracy. Furthermore, its overall retrieval performance surpasses that of other mainstream methods in these constrained scenarios. Collectively, these results establish Edge-ANN as a state-of-the-art solution for enabling large-scale, high-performance, real-time remote sensing feature retrieval on edge devices with exceptionally constrained storage. The codes of Edge-ANN are available at https://github.com/huaijiao666/Edge-ANN.

Figures (9)

• Figure 1: Comparison of the partitioning mechanisms in traditional tree-based ANN and Edge-ANN. (a) Traditional tree-based ANN employs explicitly defined and stored $d$-dimensional hyperplanes for recursive partitioning. (b) Edge-ANN implicitly defines partitions using pairs of anchor points selected from the dataset, thereby avoiding the storage of high-dimensional vectors.
• Figure 2: Edge-ANN vs. Baselie Retrieval Performance and Model Size: (a) Results of models under different primary storage constraints across three datasets; (b) Model size, with the first, second, and third rows corresponding to MillionAID, Hi-UCD, and GUN datasets, respectively. For clearer details, please refer to Material 2.
• Figure 3: Results on the MillionAID dataset: (a) Retrieval performance of Edge-ANN versus competing methods; (b) Corresponding model sizes.
• Figure 4: Results on the Hi-UCD dataset: (a) Retrieval performance of Edge-ANN versus competing methods; (b) Corresponding model sizes.
• Figure 5: Results on the GUN dataset: (a) Retrieval performance of Edge-ANN versus competing methods; (b) Corresponding model sizes.