diff --git a/src/cDNA_segmentation.py b/src/cDNA_segmentation.py
index 80a348d..878c56a 100644
--- a/src/cDNA_segmentation.py
+++ b/src/cDNA_segmentation.py
@@ -203,8 +203,9 @@ def gridding(gray: np.ndarray) -> tuple:
 # 第四部分：后处理（参考choice.m, choosemaxobj.m）
 # ============================================================
 
-def remove_small_objects(binary: np.ndarray, min_size: int = 20) -> np.ndarray:
-    """去除面积小于min_size的连通域"""
+def remove_small_objects(binary: np.ndarray, cell_area: int = 1225, pct: float = 0.02) -> np.ndarray:
+    """去除面积小于 cell_area*pct 的连通域（pct默认2%）"""
+    min_size = int(cell_area * pct)
     labeled, num = ndimage.label(binary)
     result = binary.copy()
     for i in range(1, num + 1):
@@ -213,8 +214,9 @@ def remove_small_objects(binary: np.ndarray, min_size: int = 20) -> np.ndarray:
     return result
 
 
-def keep_largest_object(binary: np.ndarray, min_size: int = 20) -> np.ndarray:
-    """只保留最大连通域"""
+def keep_largest_object(binary: np.ndarray, cell_area: int = 1225, pct: float = 0.01) -> np.ndarray:
+    """只保留最大连通域，若最大块面积 < cell_area*pct 则整块抹黑（pct默认1%）"""
+    min_size = int(cell_area * pct)
     labeled, num = ndimage.label(binary)
     if num == 0:
         return binary
@@ -373,6 +375,11 @@ def main() -> None:
 
     # ---- 步骤3: 全图逐块分割（Otsu + TV去噪） ----
     print("\n[步骤3] 全图逐块分割...")
+    # 计算每个格子的面积，用于自适应后处理阈值
+    cell_h = int(np.median(np.diff(y_grid))) if len(y_grid) > 1 else 35
+    cell_w = int(np.median(np.diff(x_grid))) if len(x_grid) > 1 else 35
+    cell_area = cell_h * cell_w
+
     bw_full = np.zeros_like(gray)
     if len(x_grid) >= 2 and len(y_grid) >= 2:
         for i in range(len(y_grid) - 1):
@@ -396,11 +403,12 @@ def main() -> None:
                     bw_blk = (blk_denoised > T).astype(np.uint8)
                 except ValueError:
                     bw_blk = np.zeros(blk.shape, dtype=np.uint8)
-                # 后处理：保留最大连通域
-                bw_blk = keep_largest_object(bw_blk, min_size=8)
+                # 后处理：保留最大连通域（最小面积 = 格子面积的1%）
+                bw_blk = keep_largest_object(bw_blk, cell_area=cell_area)
                 bw_full[r1:r2, c1:c2] = bw_blk
 
-    bw_full = remove_small_objects(bw_full, min_size=20)
+    # 全局去小连通域（最小面积 = 格子面积的2%）
+    bw_full = remove_small_objects(bw_full, cell_area=cell_area)
 
     fig_full = plot_full_segmentation(gray, bw_full, "全图逐块Otsu分割结果")
     fig_full.savefig(os.path.join(OUTPUT_DIR, 'result_full_segmentation.png'), dpi=150, bbox_inches='tight')