| 基于点云空间特征的随机森林海面光子提取方法 |
| A Random Foresl-Based Method for Extracting Sea Surface Photons Based on PointCloud Spatial Features |
| |
| DOI：10.3969/j.issn.1003-2029.2025.06.001 |
| 中文关键词: ICESal-2 海面探测 随机森林 DBSCAN |
| 英文关键词:ICESat-2 marine surrlace detection random fores DBSCAN |
| 基金项目:2024年中交集团青年创新项目（PJ2024033354）；山东省自然科学基金资助项目（ZR2024QD062） |
|
| 摘要点击次数: 98 |
| 全文下载次数: 53 |
| 中文摘要: |
| 冰、云和陆地高程卫星2号(lce,Cloud,and land Elevation Satellite-2,ICESat-2)搭载的高灵敏度光子计数激光雷达为海面探测开辟了新途径。然而，受到传感器对光子信号的极高敏感性及海气跨介质反射特性的双重约束，原始数据中表现出海面有效光子与噪声光子混杂的特点，限制了数据的直接应用。为此，本文提出了一种基于多维度空间特征挖掘的随机森林分类框架，建立从粗提取到精确分类的两級处理策略。首先基于高程的光子数量直方图的信噪比分析，实现海面光子的粗提取，继而通过人工标注构建高置信度训练数据集;提取7维特征参量训练分类模型以进一步优化分类结果。为验证方法优势，利用上述方法提取青岛近海区城日夜间数据海面光子，并与具有桑声的基于密度的聚类算法 (Density-Based Spatial Clustering of Applications with Noise， DBSCAN)进行对比分析。结果表明：DBSCAN算法虽能实现海面光子提取(搜索半径1m时准确率较高)，但其性能受参数敏感性影响，鲁棒性较低;而本文提取算法通过构建光子空间特征，在日间和夜间数据中分别取得9692%与99.75%的总体分类精度，效果优于传统方法。同时，引入高斯拟合算法提取一类水体(夏成夷拉奈岛)与二类水体(青岛近海)数据海面点并与本文方法对比，验证了本文方法在不同海城的鲁棒性。因此，本方法有效解决了强噪声背景下信号提取难题，证实机器学习策略在单光子遥感数据处理中的显著优势，为星载光子计数数据的海洋学应用提供了新的方法论支撑。 |
| 英文摘要: |
| The highly sensitive photon-counting lidar carried by the ICESat-2 (Ice, Cloud, and land Elevation Satellite-2) Satellite has opened up new avenues for sea surface detection. However, due to the dual constraints of the sensor's extremely high sensitivity to photon signals and the cross-medium reflection characteristics between the sea and the atmosphere, the original data shows the feature of a mixture of effective photons from the sea surface and noise photons, which limits the direct application of the data. To this end, this paper proposes a random forest classification framework based on multi-dimensional spatial feature mining and establishes a two-level processing strategy from coarse extraction to precise classification. Firstly, based on the signal-to-noise ratio analysis of the photon quantity histogram of elevation, the coarse extraction of sea surface photons is achieved, and then a high-confidence training dataset is constructed through manual annotation. Extract 7-dimensional feature parameters to train the classification model to further optimize the classification results. To verify the advantages of the method, the above-mentioned method was used to extract the sea surface photons of the day-night data in the coastal area of Qingdao and conduct a comparative analysis with the DBSCAN algorithm. The results show that although the DBSCAN algorithm can achieve photon extraction from the sea surface (with a relatively high accuracy rate when the search radius is 1 m), its performance is affected by parameter sensitivity and its robustness is relatively low. The extraction algorithm in this paper achieved overall classification accuracies of 96.92% and 99.75% respectively in daytime and nighttime data by constructing photon spatial features, and its performance was superior to that of traditional methods. Meanwhile, the Gaussian fitting algorithm was introduced to extract the sea surface points of Class I water bodies (Lanai Island, Hawaii) and Class II water bodies (near the coast of Qingdao), and compared with the method in this study, verifying the robustness of this method in different sea areas. Therefore, this method effectively resolves the challenge of signal extraction in a strong noise background, confirming the significant advantages of machine learning strategies in the processing of single-photon remote sensing data, and providing new methodological support for the oceanographic application of spaceborne photon counting data. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
