{"id":2145,"date":"2025-03-06T16:07:27","date_gmt":"2025-03-06T08:07:27","guid":{"rendered":"https:\/\/www.yusian.com\/blog\/?p=2145"},"modified":"2025-03-06T17:12:59","modified_gmt":"2025-03-06T09:12:59","slug":"%e7%94%a8-cnn%ef%bc%88%e5%8d%b7%e7%a7%af%e7%a5%9e%e7%bb%8f%e7%bd%91%e7%bb%9c%ef%bc%89%e8%af%86%e5%88%ab%e6%89%8b%e5%86%99%e6%95%b0%e5%ad%97%ef%bc%880-9%ef%bc%89","status":"publish","type":"post","link":"https:\/\/www.yusian.com\/blog\/article\/2025\/03\/06\/1607272145.html","title":{"rendered":"\u7528 CNN\uff08\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff09\u8bc6\u522b\u624b\u5199\u6570\u5b57\uff080-9\uff09"},"content":{"rendered":"<p>\u7b80\u5355\u793a\u4f8b\uff0c\u4e00\u4e2a\u7528 CNN \u8bc6\u522b MNIST \u624b\u5199\u6570\u5b57\u7684\u6a21\u578b\u3002\u6570\u636e\u662f 60,000 \u5f20\u8bad\u7ec3\u548c 10,000 \u5f20\u6d4b\u8bd5\u7684 28&#215;28 \u7070\u5ea6\u56fe\uff0c\u5148\u5f52\u4e00\u5316\u5230 0-1\uff0c\u518d\u52a0\u901a\u9053\u3002\u6a21\u578b\u7528\u4e24\u5c42\u5377\u79ef\uff0832 \u548c 64 \u4e2a\u6ee4\u6ce2\u5668\uff09\u63d0\u53d6\u7279\u5f81\uff0c\u4e24\u5c42\u6c60\u5316\u7f29\u5c0f\u5c3a\u5bf8\uff0c\u518d\u5c55\u5e73\u540e\u7528\u4e24\u4e2a\u5168\u8fde\u63a5\u5c42\uff08128 \u548c 10 \u4e2a\u795e\u7ecf\u5143\uff09\u8f93\u51fa\u6982\u7387\u3002\u8bad\u7ec3 5 \u8f6e\uff0c\u7528 adam \u4f18\u5316\uff0c\u635f\u5931\u662f\u4ea4\u53c9\u71b5\u3002<\/p>\n<pre><code class=\"language-python line-numbers\">import tensorflow as tf\nimport numpy as np\n\n# 1. \u52a0\u8f7dMNIST \u6570\u636e\u96c6\n(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()\n\n# 2. \u6570\u636e\u9884\u5904\u7406,\u5f52\u4e00\u5316\u3001\u52a0\u901a\u9053\u7ef4\u5ea6\nx_train = x_train \/ 255.0\nx_test = x_test \/ 255.0\nx_train = x_train.reshape(-1, 28, 28, 1)\nx_test = x_test.reshape(-1, 28, 28, 1)\n\n# 3. \u6784\u5efa\u6a21\u578b\nmodel = tf.keras.Sequential([\n    tf.keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)), # \u53bb\u8d1f\u503c\u5377\u79ef\n    tf.keras.layers.MaxPooling2D((2, 2)), # \u6c60\u5316\n    tf.keras.layers.Conv2D(64, (3, 3), activation='relu'), # 2\u6b21\u5377\u79ef\uff0c\u63d0\u53d664\u4e2a\u7279\u5f81\u503c\n    tf.keras.layers.MaxPooling2D((2, 2)), # \u6c60\u5316\n    tf.keras.layers.Flatten(), # \u5c55\u5e73\n    tf.keras.layers.Dense(128, activation='relu'), # \u5168\u8fde\u63a5\u5c42, 128\u4e2a\u795e\u7ecf\u5143\n    tf.keras.layers.Dense(10, activation='softmax') # \u8f93\u51fa\u5c42\uff0c10\u4e2a\u795e\u7ecf\u5143\n])\n\n# 4. \u7f16\u8bd1\u6a21\u578b\nmodel.compile(optimizer='adam',\n              loss='sparse_categorical_crossentropy',\n              metrics=['accuracy'])\n\n# 5. \u8bad\u7ec3\u6a21\u578b\nhistory = model.fit(x_train, y_train, epochs=5, validation_data=(x_test, y_test))\n\n# 6. \u6d4b\u8bd5\u6a21\u578b\ntest_loss, test_acc = model.evaluate(x_test, y_test)\nprint(f\"\u6d4b\u8bd5\u96c6\u635f\u5931\uff1a{test_loss:.4f}, \u6d4b\u8bd5\u96c6\u51c6\u786e\u7387\uff1a{test_acc:.4f}\")\n\n# 7. \u9884\u6d4b10\u5f20\u56fe\u7247\npredictions = np.argmax(model.predict(x_test[:10]), axis=-1)\nprint(f\"\u9884\u6d4b\u7ed3\u679c\uff1a{predictions}\")\nprint(f\"\u771f\u5b9e\u7ed3\u679c\uff1a{y_test[:10]}\")\n<\/code><\/pre>\n<p>\u8f93\u51fa\u7ed3\u679c\uff1a<\/p>\n<pre><code class=\"language-terminal line-numbers\">Epoch 1\/5\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 18s 9ms\/step - accuracy: 0.9097 - loss: 0.2911 - val_accuracy: 0.9859 - val_loss: 0.0449\nEpoch 2\/5\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 19s 10ms\/step - accuracy: 0.9872 - loss: 0.0414 - val_accuracy: 0.9889 - val_loss: 0.0338\nEpoch 3\/5\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 19s 10ms\/step - accuracy: 0.9916 - loss: 0.0260 - val_accuracy: 0.9889 - val_loss: 0.0351\nEpoch 4\/5\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 21s 11ms\/step - accuracy: 0.9937 - loss: 0.0186 - val_accuracy: 0.9914 - val_loss: 0.0268\nEpoch 5\/5\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 22s 12ms\/step - accuracy: 0.9953 - loss: 0.0142 - val_accuracy: 0.9907 - val_loss: 0.0306\n313\/313 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 1s 3ms\/step - accuracy: 0.9882 - loss: 0.0411     \n\u6d4b\u8bd5\u96c6\u635f\u5931\uff1a0.0306, \u6d4b\u8bd5\u96c6\u51c6\u786e\u7387\uff1a0.9907\n1\/1 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 0s 78ms\/step\n\u9884\u6d4b\u7ed3\u679c\uff1a[7 2 1 0 4 1 4 9 5 9]\n\u771f\u5b9e\u7ed3\u679c\uff1a[7 2 1 0 4 1 4 9 5 9]\n<\/code><\/pre>\n<p>\u624b\u5199\u56fe\u7247\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.yusian.com\/blog\/wp-content\/uploads\/2025\/03\/Figure_1-1.png\" alt=\"\" \/><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7b80\u5355\u793a\u4f8b\uff0c\u4e00\u4e2a\u7528 CNN \u8bc6\u522b MNIST \u624b\u5199\u6570\u5b57\u7684\u6a21\u578b\u3002\u6570\u636e\u662f 60,000 \u5f20\u8bad\u7ec3\u548c 10,000 \u5f20\u6d4b\u8bd5\u7684 28&#215;28 \u7070\u5ea6\u56fe\uff0c\u5148\u5f52\u4e00\u5316\u5230 0-1\uff0c\u518d\u52a0\u901a\u9053\u3002\u6a21\u578b\u7528\u4e24\u5c42\u5377\u79ef\uff0832 \u548c 64 \u4e2a\u6ee4\u6ce2\u5668\uff09\u63d0\u53d6\u7279\u5f81\uff0c\u4e24\u5c42\u6c60\u5316\u7f29\u5c0f\u5c3a\u5bf8\uff0c\u518d\u5c55\u5e73\u540e\u7528\u4e24\u4e2a\u5168\u8fde\u63a5\u5c42\uff08128 \u548c 10 \u4e2a\u795e\u7ecf\u5143\uff09\u8f93\u51fa\u6982\u7387\u3002\u8bad\u7ec3 5 \u8f6e\uff0c\u7528 adam \u4f18\u5316\uff0c\u635f\u5931\u662f\u4ea4\u53c9\u71b5\u3002 import tensorflow as tf import numpy as np # 1. \u52a0\u8f7dMNIST \u6570\u636e\u96c6 (x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data() # 2. \u6570\u636e\u9884\u5904\u7406,\u5f52\u4e00\u5316\u3001\u52a0\u901a\u9053\u7ef4\u5ea6 x_train = x_train \/ 255.0 x_test = x_test \/ 255.0 x_train = [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[416,469,475],"class_list":["post-2145","post","type-post","status-publish","format-standard","hentry","category-article","tag-ai","tag-cnn","tag-mnist"],"_links":{"self":[{"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/posts\/2145","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/comments?post=2145"}],"version-history":[{"count":0,"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/posts\/2145\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/media?parent=2145"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/categories?post=2145"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.yusian.com\/blog\/wp-json\/wp\/v2\/tags?post=2145"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}