# SG filter function
def apply_sg_filter(collection, band_names, window_size, order, roi):
    # 自动计算起始日期
    start_date = ee.Date(collection.sort('system:time_start').first().get('system:time_start'))
    def prepare_image(img):
        dstamp = ee.Date(img.get('system:time_start'))
        ddiff = dstamp.difference(start_date, 'day')
        const_bands = ee.Image.constant([1, ddiff, ddiff.pow(2), ddiff.pow(3)]) \
                        .rename(['constant', 't', 't2', 't3']) \
                        .float()
        return img.select(band_names).addBands(const_bands) \
                  .set('date', dstamp)
    
    prepared_collection = collection.map(prepare_image)
    
    # 设置 SG 滤波参数
    half_window = (window_size - 1) // 2
    image_axis = 0
    band_axis = 1
    coeff_flattener = [['constant', 'x', 'x2', 'x3']]
    indep_selectors = ['constant', 't', 't2', 't3']
    
    def get_local_fit(i, band_name, geometry=roi):
        # 将选定的波段和独立变量转换为数组
        array = prepared_collection.select([band_name] + indep_selectors).toArray()  # 返回类型: Array
        # 提取当前图像及其窗口内的图像
        subarray = array.arraySlice(image_axis, ee.Number(i).int(), ee.Number(i).add(window_size).int())  # 返回类型: Array
        # 分离独立变量和响应变量
        predictors = subarray.arraySlice(band_axis, 1, order + 2)  # 独立变量，返回类型: Array
        response = subarray.arraySlice(band_axis, 0, 1)  # 响应变量，返回类型: Array
    
        # 输出维度信息
        predictors_shape = predictors.arrayLength(0)  # 获取行数
        predictors_band_count = predictors.arrayLength(1)  # 获取列数
        response_shape = response.arrayLength(0)  # 行数（应与 predictors 行数相同）
    
        # 检查 NaN 和无效值
        predictors_image = ee.Image(predictors)  # 将 Array 转换为 Image
        has_nan_predictors = predictors_image.updateMask(predictors_image.eq(float('nan'))).reduceRegion(
            reducer=ee.Reducer.anyNonZero(),
            geometry=geometry,  # 使用传入的 geometry
            scale=30  # 根据需要调整比例
        ).get('array')
    
        response_image = ee.Image(response)  # 将 Array 转换为 Image
        has_nan_response = response_image.updateMask(response_image.eq(float('nan'))).reduceRegion(
            reducer=ee.Reducer.anyNonZero(),
            geometry=geometry,  # 使用传入的 geometry
            scale=30  # 根据需要调整比例
        ).get('array')
    
        # 记录检查结果
        log_info = {
            'Predictors shape': predictors_shape,
            'Predictors band count': predictors_band_count,
            'Response shape': response_shape,
            'Has NaN in predictors': has_nan_predictors,
            'Has NaN in response': has_nan_response
        }
        
        # 使用 ee.Algorithms.Log 来输出调试信息
        for key, value in log_info.items():
            print(key + ': ' + str(value))
    
        # 通过线性代数计算拟合系数
        coeff = predictors.matrixSolve(response)  # 返回类型: Array
        # 返回扁平化的系数
        return coeff.arrayProject([0]).arrayFlatten(coeff_flattener)  # 返回类型: Array

    
    # 应用 SG 滤波
    collection_size = prepared_collection.size()
    run_length = ee.List.sequence(0, collection_size.subtract(window_size))
    
    def smooth_band(band_name):
        def apply_sg(i):
            ref = prepared_collection.toList(collection_size).get(ee.Number(i).add(half_window))
            return get_local_fit(i, band_name).multiply(ee.Image(ref).select(indep_selectors)) \
                               .reduce(ee.Reducer.sum()) \
                               .rename(band_name) \
                               .copyProperties(ee.Image(ref), ['system:time_start', 'system:index']) \
                               .set('date', ee.Date(ee.Image(ref).get('system:time_start')))
        
        return ee.ImageCollection(run_length.map(apply_sg))
    
    # 对每个波段应用 SG 滤波
    smoothed_collections = [smooth_band(band) for band in band_names]
    
    # 合并所有平滑后的波段
    smoothed_collection = ee.ImageCollection(smoothed_collections[0])
    for col in smoothed_collections[1:]:
        smoothed_collection = smoothed_collection.combine(ee.ImageCollection(col))
    
    # 应用掩膜到每个平滑后的影像
    masked_smoothed_collection = smoothed_collection.map(lambda image: image.clip(roi))
    
    return masked_smoothed_collection.sort('system:time_start') 

images = collection.toList(collection.size())

size = 37
for i in range(size):
    image = ee.Image(images.get(i))
    
    outname = assetid+r'/'+out_dir.split(os.sep)[-1]+r"_Season"+str(i+1)
    # Export Data
    geemap.ee_export_image_to_asset(image,
                                    description=outname.split('/')[-1],
                                    assetId=outname, #projects/ee-xihuangrsvp/assets/QiangTang  r'projects/ee-igsnrr2024/assets/'+outname
                                    region=block_roi,
                                    crs='epsg:4326',
                                    scale=30)