High-Fidelity Lake Extraction via Two-Stage Prompt Enhancement: Establishing a Novel Baseline and Benchmark

TL;DR

The paper tackles the challenge of robust lake extraction from remote sensing imagery, where diverse lake morphologies and data noise hinder accurate segmentation. It introduces LEPrompter, a two-stage prompt enhancement framework that guides training with a lightweight prompt encoder and decoder while enabling prompt-free inference, and it constructs a prompt-based benchmark dataset using morphological operations and DBSCAN to generate point, box, and mask prompts. The study demonstrates state-of-the-art or near-SOTA improvements on SW and QTPL datasets, with modest parameter and FLOP overhead during the prompt-based stage and zero-cost inference. The work provides a practical baseline for automated lake extraction and offers a principled approach to integrating prompts into semantic segmentation for remote sensing, with potential applicability to broader image analysis tasks.

Abstract

Lake extraction from remote sensing imagery is a complex challenge due to the varied lake shapes and data noise. Current methods rely on multispectral image datasets, making it challenging to learn lake features accurately from pixel arrangements. This, in turn, affects model learning and the creation of accurate segmentation masks. This paper introduces a prompt-based dataset construction approach that provides approximate lake locations using point, box, and mask prompts. We also propose a two-stage prompt enhancement framework, LEPrompter, with prompt-based and prompt-free stages during training. The prompt-based stage employs a prompt encoder to extract prior information, integrating prompt tokens and image embedding through self- and cross-attention in the prompt decoder. Prompts are deactivated to ensure independence during inference, enabling automated lake extraction without introducing additional parameters and GFlops. Extensive experiments showcase performance improvements of our proposed approach compared to the previous state-of-the-art method. The source code is available at https://github.com/BastianChen/LEPrompter.

Figures (6)

• Figure 1: Our proposed two-stage prompt enhancement framework for lake extraction. The prompt-based approach simulates a teacher guiding students to solve challenging problems, while the prompt-free approach allows students to tackle problems independently. Conversely, the inference process exclusively utilizes the prompt-free approach.
• Figure 2: Visualization images of our proposed benchmark.
• Figure 3: The workflow diagram of creating our benchmark for lake extraction.
• Figure 4: Overview architecture of our proposed enhancement framework LEPrompter and lightweight prompt decoder.
• Figure 5: (a) Influence of the prompt-based steps. (b) Influence of the type and number of prompt points on our approach.