Surficial sediment sampling is a necessary operation in deep-sea sediment studies. However, the current approach to select sampling locations lacks clarity and definitive guidelines. To solve this problem, we propose an innovative classification-based guided sampling method. Our method leverages the self-developed high-frequency submersible subbottom profiler (HF-SSBP) to capture the high-resolution SBP images of surficial sediments. Building upon the convolutional neural network—bidirectional long short-term memory (CNN-BLSTM) model, we fuse acoustic faces specific to high-resolution SBP images for classification purposes. The classification results effectively distinguish the various sediment structures, particularly with minor variations. The precision of the model exceeds 95%. We can significantly reduce redundancy in sampling similar sedimentary structures by implementing this method. This approach allows us to maximize the collection of high-value samples in the face of limited sampling conditions. We validated the feasibility of the approach using high-resolution SBP data of surficial sediments in the South China Sea continental slope.

中文翻译：

---

使用声学面 CNN-BLSTM 融合指导深海表层沉积物采样

表层沉积物采样是深海沉积物研究中的必要操作。然而，目前选择采样地点的方法缺乏明确性和明确的指导方针。为了解决这个问题，我们提出了一种创新的基于分类的引导采样方法。我们的方法利用自主开发的高频潜水式海底剖面仪（HF-SSBP）来捕获表层沉积物的高分辨率SBP图像。基于卷积神经网络双向长短期记忆 (CNN-BLSTM) 模型，我们融合特定于高分辨率 SBP 图像的声学面孔以进行分类。分类结果有效区分了各种沉积物结构，特别是变化较小的沉积物结构。模型精度超过95%。通过实施这种方法，我们可以显着减少对类似沉积结构进行采样的冗余。这种方法使我们能够在有限的采样条件下最大限度地收集高价值样本。我们利用南海大陆坡表层沉积物的高分辨率 SBP 数据验证了该方法的可行性。