在海洋数据处理中，不可避免地会遇到NAN值的处理，当全部将NAN替换成“0”进行处理，有时候的结果可能不准确。这里以Matlab为例，对流场数据进行滤波，这里采用smooth2a、中值滤波两种方法进行对比：
中值滤波的结果，窗口7x7
smooth2a的结果，窗口7x7
可以看到smooth2a的效果很不理想，而中值滤波在边界处呈现白边。
为了解决这个问题，我们自己定义一个针对包含nan数组（矩阵）的滤波函数slidingWindowFilter。
function filteredMatrix = slidingWindowFilter(matrix, windowSize, filterType)[rows, cols] = size(matrix);filteredMatrix = NaN(rows, cols);halfWindowSize = floor(windowSize / 2);for i = 1:rowsfor j = 1:colsr1 = max(1, i - halfWindowSize);r2 = min(rows, i + halfWindowSize);c1 = max(1, j - halfWindowSize);c2 = min(cols, j + halfWindowSize);window = matrix(r1:r2, c1:c2);if all(~isnan(window(:)))switch filterTypecase 'mean'filteredMatrix(i, j) = mean(window(:));case 'median'filteredMatrix(i, j) = median(window(:));case 'gaussian'sigma = 1;[X, Y] = meshgrid(-halfWindowSize:halfWindowSize, -halfWindowSize:halfWindowSize);G = exp(-(X.^2 + Y.^2) / (2*sigma^2));G = G / sum(G(:));filteredMatrix(i, j) = sum(sum(window .* G));case 'max'filteredMatrix(i, j) = max(window(:));case 'min'filteredMatrix(i, j) = min(window(:));case 'laplacian'LaplacianKernel = [1 -2 1; -2 4 -2; 1 -2 1];filteredMatrix(i, j) = sum(sum(window .* LaplacianKernel));case 'rms'filteredMatrix(i, j) = sqrt(mean(window(:).^2));case 'bilateral'sigma_spatial = 1;sigma_intensity = 1;[X, Y] = meshgrid(-halfWindowSize:halfWindowSize, -halfWindowSize:halfWindowSize);spatialWeights = exp(-(X.^2 + Y.^2) / (2*sigma_spatial^2));intensityWeights = exp(-(window - matrix(i, j)).^2 / (2*sigma_intensity^2));bilateralWeights = spatialWeights .* intensityWeights;bilateralWeights = bilateralWeights / sum(bilateralWeights(:));filteredMatrix(i, j) = sum(sum(window .* bilateralWeights));case 'adaptive'localVariance = var(window(:));globalVariance = var(matrix(:));adaptiveFactor = globalVariance / (globalVariance + localVariance);filteredMatrix(i, j) = adaptiveFactor * mean(window(:)) + ...(1 - adaptiveFactor) * matrix(i, j);otherwiseerror('Unknown filter type.');endendendendend
调用示例：
% 使用拉普拉斯滤波filteredMatrix = slidingWindowFilter(matrix, windowSize, 'laplacian');
结果：
想要效果更好，可以再加入其他滤波进行处理。
