三维空间中堆状数据集的边界分离方法技术问询

针对你描述的堆状分布点集，完全可以构建近似边界实现与空白区域的分离，以下是几种实用方案：

一、密度聚类+凸包边界（图形化表示） #

先通过DBSCAN识别独立的点簇，再为每个簇计算凸包作为边界：

% 假设XYZ是n×3的点集矩阵
eps_val = 0.5; % 邻域半径，需根据点集密度调整
min_points = 5; % 邻域内最小点数，控制聚类粒度
idx = dbscan(XYZ, eps_val, min_points);

% 可视化各簇凸包边界
figure; hold on;
valid_clusters = unique(idx(idx~=-1)); % 过滤噪声点
color_map = jet(length(valid_clusters));

for i = 1:length(valid_clusters)
    cluster_data = XYZ(idx==valid_clusters(i), :);
    hull_faces = convhulln(cluster_data); % 计算三维凸包
    trisurf(hull_faces, cluster_data(:,1), cluster_data(:,2), cluster_data(:,3), ...
        'FaceColor', color_map(i,:), 'FaceAlpha', 0.3, 'EdgeColor', 'k');
end

plot3(XYZ(:,1), XYZ(:,2), XYZ(:,3), 'o', 'MarkerSize', 2);
grid on; daspect([1 1 1]);

凸包会贴合每个点簇的外围，直观分隔簇与空白区域。

二、核密度估计（KDE）边界（解析+图形化） #

通过KDE生成点集的密度分布，以密度阈值定义边界：

% 生成三维网格
grid_res = 50; % 网格分辨率，越高边界越精细
[X_grid,Y_grid,Z_grid] = meshgrid(...
    linspace(min(XYZ(:,1)), max(XYZ(:,1)), grid_res), ...
    linspace(min(XYZ(:,2)), max(XYZ(:,2)), grid_res), ...
    linspace(min(XYZ(:,3)), max(XYZ(:,3)), grid_res));
grid_points = [X_grid(:), Y_grid(:), Z_grid(:)];

% 计算核密度（需Statistics Toolbox）
kde_vals = ksdensity(XYZ, grid_points, 'Function', 'pdf');
kde_vals = reshape(kde_vals, size(X_grid));

% 绘制密度等高面作为边界
figure; hold on;
density_threshold = 0.01; % 阈值需根据实际密度调整
contour3(X_grid,Y_grid,Z_grid, kde_vals, [density_threshold density_threshold], 'LineWidth', 2);
plot3(XYZ(:,1), XYZ(:,2), XYZ(:,3), 'o', 'MarkerSize', 2);
grid on; daspect([1 1 1]);

解析上可定义“密度大于density_threshold的区域为簇”，同时等高面直观展示边界。

三、支持向量机（SVM）分类边界（解析描述） #

若能标注少量簇内/空白区域样本，可训练SVM得到精确的分隔边界：

% 假设已标注标签：label为n×1向量，1=簇内点，0=空白区域点
svm_model = fitcsvm(XYZ, label);

% 生成网格并预测区域类别
grid_res = 30;
[X_grid,Y_grid,Z_grid] = meshgrid(...
    linspace(min(XYZ(:,1)), max(XYZ(:,1)), grid_res), ...
    linspace(min(XYZ(:,2)), max(XYZ(:,2)), grid_res), ...
    linspace(min(XYZ(:,3)), max(XYZ(:,3)), grid_res));
grid_points = [X_grid(:), Y_grid(:), Z_grid(:)];
pred_labels = predict(svm_model, grid_points);
pred_labels = reshape(pred_labels, size(X_grid));

% 绘制SVM决策边界
figure; hold on;
isosurface(X_grid,Y_grid,Z_grid, pred_labels, 0.5);
plot3(XYZ(label==1,1), XYZ(label==1,2), XYZ(label==1,3), 'o', 'MarkerSize', 2);
grid on; daspect([1 1 1]);

SVM的决策边界基于超平面组合，可通过模型参数进行解析描述，适合需要数学定义边界的场景。

内容的提问来源于stack exchange，提问作者Andy Ayr