180 lines
6.4 KiB
Python
180 lines
6.4 KiB
Python
from ultralytics import YOLO
|
||
import cv2
|
||
import numpy as np
|
||
import os
|
||
|
||
# ------------------ 配置 ------------------
|
||
model_path = 'ultralytics_yolo11-main/runs/train/exp4/weights/best.pt'
|
||
img_folder = '/home/hx/yolo/test' # 输入文件夹路径
|
||
output_mask_dir = 'output_masks'
|
||
os.makedirs(output_mask_dir, exist_ok=True)
|
||
|
||
SUPPORTED_FORMATS = ('.jpg', '.jpeg', '.png', '.bmp', '.tif', '.tiff')
|
||
|
||
# ------------------ 加载模型 ------------------
|
||
model = YOLO(model_path)
|
||
model.to('cuda') # 使用 GPU
|
||
|
||
def vector_to_normal(v):
|
||
"""从方向向量得到逆时针旋转90°的法向量"""
|
||
return np.array([-v[1], v[0]])
|
||
|
||
def angle_between_normals(n1, n2):
|
||
"""计算两个法向量之间的最小夹角 [0°, 90°]"""
|
||
n1_u = n1 / (np.linalg.norm(n1) + 1e-8)
|
||
n2_u = n2 / (np.linalg.norm(n2) + 1e-8)
|
||
cos_theta = np.clip(np.dot(n1_u, n2_u), -1.0, 1.0)
|
||
angle_deg = np.degrees(np.arccos(cos_theta))
|
||
return min(angle_deg, 180 - angle_deg)
|
||
|
||
def process_image(img_path, output_dir):
|
||
img = cv2.imread(img_path)
|
||
if img is None:
|
||
print(f"❌ 无法读取图像: {img_path}")
|
||
return
|
||
|
||
h, w = img.shape[:2]
|
||
filename = os.path.basename(img_path)
|
||
name_only = os.path.splitext(filename)[0]
|
||
|
||
print(f"\n🔄 正在处理: {name_only}")
|
||
|
||
# ------------------ 生成合成掩码 ------------------
|
||
composite_mask = np.zeros((h, w), dtype=np.uint8)
|
||
results = model(img_path, imgsz=1280, conf=0.5)
|
||
|
||
for r in results:
|
||
if r.masks is not None:
|
||
masks = r.masks.data.cpu().numpy()
|
||
for mask in masks:
|
||
mask_resized = cv2.resize(mask, (w, h))
|
||
mask_img = (mask_resized * 255).astype(np.uint8)
|
||
composite_mask = np.maximum(composite_mask, mask_img)
|
||
|
||
mask_save_path = os.path.join(output_dir, f'mask_{name_only}.png')
|
||
cv2.imwrite(mask_save_path, composite_mask)
|
||
print(f"✅ 掩码已保存: {mask_save_path}")
|
||
|
||
# ------------------ 提取轮廓 ------------------
|
||
contours, _ = cv2.findContours(composite_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
|
||
if len(contours) == 0:
|
||
print(f"⚠️ 未检测到轮廓: {filename}")
|
||
return
|
||
|
||
all_contours = np.vstack(contours)
|
||
epsilon = 0.005 * cv2.arcLength(all_contours, True)
|
||
approx = cv2.approxPolyDP(all_contours, epsilon, True)
|
||
pts = [p[0] for p in approx]
|
||
|
||
# ------------------ 提取前4条最长边 ------------------
|
||
line_segments = []
|
||
n = len(pts)
|
||
for i in range(n):
|
||
p1 = np.array(pts[i])
|
||
p2 = np.array(pts[(i + 1) % n])
|
||
length = np.linalg.norm(p2 - p1)
|
||
if length > 20:
|
||
line_segments.append({
|
||
'index': i,
|
||
'start': p1,
|
||
'end': p2,
|
||
'length': length,
|
||
'vector': p2 - p1
|
||
})
|
||
|
||
# 按长度排序，取前4条
|
||
line_segments.sort(key=lambda x: x['length'], reverse=True)
|
||
top4 = line_segments[:4]
|
||
|
||
if len(top4) < 4:
|
||
print(f"⚠️ 不足4条有效边（共{len(top4)}条）: {filename}")
|
||
return
|
||
|
||
print(f"✅ 已提取前4条最长边")
|
||
|
||
# ------------------ 计算每条边的法向量 ------------------
|
||
normals = []
|
||
for edge in top4:
|
||
v = edge['vector']
|
||
n = vector_to_normal(v)
|
||
normals.append(n)
|
||
|
||
# ------------------ 计算所有法向量之间的夹角，找最小夹角对 ------------------
|
||
min_angle = float('inf')
|
||
remove_pair = (0, 1) # 默认删除前两条
|
||
|
||
for i in range(4):
|
||
for j in range(i + 1, 4):
|
||
angle = angle_between_normals(normals[i], normals[j])
|
||
if angle < min_angle:
|
||
min_angle = angle
|
||
remove_pair = (i, j)
|
||
|
||
print(f"✅ 法向量最小夹角: {min_angle:.2f}° (边{i+1} vs 边{j+1})")
|
||
|
||
# ------------------ 删除夹角最小的一对边中的两条，保留另外两条 ------------------
|
||
keep_indices = set(range(4)) - set(remove_pair)
|
||
kept_edges = [top4[i] for i in keep_indices]
|
||
|
||
print(f"✅ 保留边索引: {sorted(keep_indices)}")
|
||
print(f"✅ 删除边索引: {remove_pair}")
|
||
|
||
if len(kept_edges) != 2:
|
||
print("❌ 保留边数量异常！")
|
||
return
|
||
|
||
edge1, edge2 = kept_edges
|
||
|
||
# ------------------ 计算两条保留边之间的直线夹角 ------------------
|
||
v1 = edge1['vector']
|
||
v2 = edge2['vector']
|
||
v1_u = v1 / (np.linalg.norm(v1) + 1e-8)
|
||
v2_u = v2 / (np.linalg.norm(v2) + 1e-8)
|
||
cos_theta = np.clip(np.dot(v1_u, v2_u), -1.0, 1.0)
|
||
raw_angle = np.degrees(np.arccos(cos_theta))
|
||
line_angle = min(raw_angle, 180 - raw_angle)
|
||
|
||
# ------------------ 打印信息 ------------------
|
||
print(f"\n✅ 保留的两条边:")
|
||
for i, e in enumerate([edge1, edge2], 1):
|
||
print(f" 边{i}: 起点{tuple(e['start'])} → 终点{tuple(e['end'])}, 长度: {e['length']:.1f}")
|
||
|
||
print(f"\n✅ 两条保留边之间的夹角: {line_angle:.2f}°")
|
||
|
||
# ------------------ 可视化 ------------------
|
||
vis_img = img.copy()
|
||
color_keep = [(0, 255, 0), (255, 0, 255)] # 绿色和品红表示保留边
|
||
|
||
for idx, e in enumerate(kept_edges):
|
||
p1, p2 = e['start'], e['end']
|
||
cv2.line(vis_img, tuple(p1), tuple(p2), color_keep[idx], 3, cv2.LINE_AA)
|
||
mid = ((p1[0] + p2[0]) // 2, (p1[1] + p2[1]) // 2)
|
||
cv2.putText(vis_img, f"{e['length']:.0f}", (mid[0], mid[1] + (idx*20)),
|
||
cv2.FONT_HERSHEY_SIMPLEX, 0.7, color_keep[idx], 2)
|
||
|
||
# 标注最终夹角
|
||
cv2.putText(vis_img, f"Final Angle: {line_angle:.2f}°", (20, 50),
|
||
cv2.FONT_HERSHEY_SIMPLEX, 1.0, (0, 255, 255), 2)
|
||
|
||
# 保存结果
|
||
vis_save_path = os.path.join(output_dir, f'kept_edges_{name_only}.jpg')
|
||
cv2.imwrite(vis_save_path, vis_img)
|
||
print(f"✅ 保留边可视化已保存: {vis_save_path}")
|
||
|
||
# ------------------ 主程序 ------------------
|
||
if __name__ == '__main__':
|
||
if not os.path.isdir(img_folder):
|
||
print(f"❌ 图像文件夹不存在: {img_folder}")
|
||
exit()
|
||
|
||
image_files = [f for f in os.listdir(img_folder) if f.lower().endswith(SUPPORTED_FORMATS)]
|
||
if len(image_files) == 0:
|
||
print(f"⚠️ 未找到支持的图像文件: {img_folder}")
|
||
exit()
|
||
|
||
print(f"✅ 发现 {len(image_files)} 张图像，开始处理...")
|
||
|
||
for file in image_files:
|
||
process_image(os.path.join(img_folder, file), output_mask_dir)
|
||
|
||
print(f"\n🎉 所有图像处理完成！结果保存在: {output_mask_dir}") |