检查此段代码，看是否还有什么问题：""" 改进版CNN人脸识别系统 包含数据增强、模型训练、特征比对完整流程 使用LFW数据集进行训练 """ import warnings import logging import os import numpy as np import tensorflow as tf from tensorflow.keras import layers, models, applications from sklearn.model_selection import train_test_split from sklearn.metrics import accuracy_score import cv2 import matplotlib.pyplot as plt # 配置环境 warnings.filterwarnings("ignore") logging.getLogger("tensorflow").setLevel(logging.ERROR) os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' # 人脸检测器 face_cascade = cv2.CascadeClassifier( cv2.data.haarcascades + 'haarcascade_frontalface_default.xml') # 超参数配置 CONFIG = { "input_shape": (160, 160, 3), # 输入尺寸 "embedding_dim": 128, # 特征向量维度 "batch_size": 32, # 训练批次大小 "epochs": 30, # 训练轮次 "data_path": "./lfw", # LFW数据集路径 "augmentation": True, # 启用数据增强 "test_size": 0.2, # 验证集比例 "threshold": 0.85 # 相似度阈值 } class FaceRecognitionSystem: def __init__(self): self.model = None self.history = None def _build_model(self): """构建改进的CNN模型""" base_model = applications.MobileNetV2( input_shape=CONFIG["input_shape"], include_top=False, weights='imagenet' ) base_model.trainable = True inputs = tf.keras.Input(shape=CONFIG["input_shape"]) x = base_model(inputs, training=True) x = layers.GlobalAveragePooling2D()(x) x = layers.Dense(256, activation='relu')(x) x = layers.Dropout(0.5)(x) outputs = layers.Dense(CONFIG["embedding_dim"])(x) self.model = tf.keras.Model(inputs, outputs) def _preprocess_image(self, img_path): """图像预处理流程""" # 读取并检测人脸 img = cv2.imread(img_path) gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) faces = face_cascade.detectMultiScale(gray, 1.3, 5) if len(faces) == 0: return None # 裁剪人脸区域 (x,y,w,h) = faces[0] face = img[y:y+h, x:x+w] # 调整尺寸和归一化 face = cv2.resize(face, CONFIG["input_shape"][:2]) face = face.astype("float32") / 255.0 # 数据增强 if CONFIG["augmentation"]: face = self._augment_image(face) return face def _augment_image(self, img): """数据增强方法""" # 随机水平翻转 if np.random.rand() > 0.5: img = cv2.flip(img, 1) # 随机旋转 angle = np.random.uniform(-15, 15) rows, cols = img.shape[:2] M = cv2.getRotationMatrix2D((cols/2, rows/2), angle, 1) img = cv2.warpAffine(img, M, (cols, rows)) # 随机亮度调整 img = cv2.convertScaleAbs(img, alpha=np.random.uniform(0.8, 1.2), beta=np.random.randint(-10, 10)) return img def _load_dataset(self): """加载LFW数据集""" image_paths = [] labels = [] # 确保路径存在 data_path = CONFIG[r"D:\BaiduNetdiskDownload\lfw"] if not os.path.exists(data_path): raise FileNotFoundError(f"数据集路径不存在: {data_path}") # 遍历所有人名目录 for person_name in os.listdir(data_path): person_dir = os.path.join(data_path, person_name) # 过滤非目录文件 if not os.path.isdir(person_dir): continue # 遍历该人目录下的所有图片 for img_name in os.listdir(person_dir): # 过滤非图片文件（可选） if not img_name.lower().endswith(('.png', '.jpg', '.jpeg')): continue img_path = os.path.join(person_dir, img_name) image_paths.append(img_path) labels.append(person_name) # 过滤无效图片并预处理 valid_images = [] valid_labels = [] for path, label in zip(image_paths, labels): try: img = self._preprocess_image(path) if img is not None: valid_images.append(img) valid_labels.append(label) except Exception as e: print(f"处理图片 {path} 时出错: {e}") # 将标签转换为整数编码 from sklearn.preprocessing import LabelEncoder le = LabelEncoder() encoded_labels = le.fit_transform(valid_labels) return np.array(valid_images), encoded_labels def _preprocess_image(self, path): """图片预处理函数""" import cv2 # 读取图片（包括中文路径处理） img = cv2.imdecode(np.fromfile(path, dtype=np.uint8), cv2.IMREAD_COLOR) if img is None: print(f"警告: 无法读取图片 {path}") return None # 调整图片尺寸（示例使用224x224） img = cv2.resize(img, (224, 224)) # 转换颜色通道（BGR -> RGB） img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # 归一化处理 img = img.astype(np.float32) / 255.0 return img def _create_pairs(self, images, labels): """创建训练正负样本对""" num_classes = len(np.unique(labels)) digit_indices = [np.where(labels == i)[0] for i in range(num_classes)] pairs = [] labels = [] # 创建正样本对 for idx in range(len(images)): _, label = images[idx], labels[idx] indices = digit_indices[label] if len(indices) >= 2: n = np.random.choice(indices) while n == idx: n = np.random.choice(indices) pairs += [[images[idx], images[n]]] labels += [1] # 创建负样本对 for idx in range(len(images)): _, label = images[idx], labels[idx] indices = np.where(labels != label)[0] n = np.random.choice(indices) pairs += [[images[idx], images[n]]] labels += [0] return np.array(pairs), np.array(labels) def train(self): """训练流程""" # 加载数据 images, labels = self._load_dataset() # 创建样本对 pairs, pair_labels = self._create_pairs(images, labels) # 划分数据集 X_train, X_val, y_train, y_val = train_test_split( pairs, pair_labels, test_size=CONFIG["test_size"]) # 构建模型 self._build_model() # 定义对比损失 def contrastive_loss(y_true, y_pred): margin = 1 square_pred = tf.square(y_pred) margin_square = tf.square(tf.maximum(margin - y_pred, 0)) return tf.reduce_mean( y_true * square_pred + (1 - y_true) * margin_square) # 编译模型 self.model.compile( optimizer=tf.keras.optimizers.Adam(0.001), loss=contrastive_loss, metrics=["accuracy"] ) # 训练模型 self.history = self.model.fit( X_train, y_train, validation_data=(X_val, y_val), batch_size=CONFIG["batch_size"], epochs=CONFIG["epochs"] ) # 保存模型 self.model.save("face_recognition_cnn.h5") def compute_similarity(self, img1_path, img2_path): """计算两张图片的相似度""" # 预处理图片 img1 = self._preprocess_image(img1_path) img2 = self._preprocess_image(img2_path) if img1 is None or img2 is None: return 0.0 # 提取特征 emb1 = self.model.predict(np.expand_dims(img1, axis=0))[0] emb2 = self.model.predict(np.expand_dims(img2, axis=0))[0] # 计算余弦相似度 similarity = np.dot(emb1, emb2) / (np.linalg.norm(emb1) * np.linalg.norm(emb2)) return similarity def verify(self, img1_path, img2_path): """验证是否为同一人""" similarity = self.compute_similarity(img1_path, img2_path) return similarity > CONFIG["threshold"], similarity def visualize_training(self): """可视化训练过程""" plt.figure(figsize=(12, 6)) plt.subplot(1, 2, 1) plt.plot(self.history.history['loss'], label='Training Loss') plt.plot(self.history.history['val_loss'], label='Validation Loss') plt.legend() plt.title('Loss Evolution') plt.subplot(1, 2, 2) plt.plot(self.history.history['accuracy'], label='Training Accuracy') plt.plot(self.history.history['val_accuracy'], label='Validation Accuracy') plt.legend() plt.title('Accuracy Evolution') plt.show() # # 使用示例 # if __name__ == "__main__": # system = FaceRecognitionSystem() # # # 训练模型（首次运行需要） # # system.train() # # system.visualize_training() # # # 加载预训练模型 # system.model = tf.keras.models.load_model("face_recognition_cnn.h5") # # # 测试验证 # is_same, similarity = system.verify("person1.jpg", "person2.jpg") # print(f"是否为同一人: {is_same}, 相似度: {similarity:.2%}")