当前位置:   article > 正文

2024年泰迪杯数据挖掘B题详细思路代码文章教程_泰迪杯2024数据

泰迪杯2024数据

目前b题已全部更新包含详细的代码模型和文章,本文也给出了结果展示和使用模型说明。

同时文章最下方包含详细的视频教学获取方式,手把手保姆级,模型高精度,结果有保障!

分析:
本题待解决问题

目标:利用提供的数据集,通过特征提取和多模态特征融合模型建立,实现图像与文本间的互检索。

具体任务:

基于图像检索的文本:利用提供的文本信息,对图像进行检索,输出相似度较高的前五张图像。

基于文本检索的图像:利用提供的图像ID,对文本进行检索,输出相似度较高的前五条文本。

数据集和任务要求

附件1:包含五万张图像和对应的文本信息。

附件2和附件3:分别提供了任务1和任务2的数据信息,包括测试集文本、图像ID和图像数据库。

附件4:提供了任务结果的模板文件。

评价标准

使用**召回率Recall at K(R@K)**作为评价指标,即查询结果中真实结果排序在前K的比率,本赛题设定K=5,即评价标准为R@5。

步骤一:构建图文检索模型

采用图文检索领域已经封装好的模型:多模态图文互检模型

基于本题附件一所给的数据进行调优

可以给大家展示以下我们模型的效果,和那种一两天做出来的效果完全不一样,我们的模型效果和两个任务的预测情况完整是准确且符合逻辑的。

在这里插入图片描述

任务一结果展示:

在这里插入图片描述

在这里插入图片描述

任务二结果展示:

在这里插入图片描述

步骤二:基于图像检索文本

1.数据预处理和特征提取

文本数据预处理:

清洗文本:去除文本中的停用词、标点符号等无关信息。

文本向量化:利用NLP技术(如Word2Vec, GloVe, BERT等)将文本转换为数值向量,以便进行计算和比较。

在这里插入图片描述

import jieba
import pandas as pd
from collections import Counter
#读取CSV文件
image_word_data = pd.read_csv('附件1/ImageWordData.csv')
#加载自定义的停用词表(如果有的话),或使用jieba内置的停用词表
#例如: stop_words = set(open('path_to_stop_words.txt').read().strip().split('\n'))
stop_words = set() # 假设暂时没有自定义的停用词表
#文本预处理函数
def preprocess_text(captions):
preprocessed_captions = []
for caption in captions:
# 使用jieba进行分词
tokens = jieba.lcut(caption)
# 去除停用词
tokens = [token for token in tokens if token not in stop_words and len(token) > 1]
# 将处理过的词加入结果列表
preprocessed_captions.append(" ".join(tokens))
return preprocessed_captions
#对caption列进行预处理
preprocessed_captions = preprocess_text(image_word_data['caption'])
#查看处理过的一些示例文本
for i in range(5):
print(preprocessed_captions[i])
#(可选)统计词频
word_counts = Counter(" ".join(preprocessed_captions).split())
print(word_counts.most_common(10))
​
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28

图像数据预处理:

图像标准化:将所有图像调整到相同的大小和色彩空间。

特征提取:使用深度学习模型(如CNN, ResNet, VGG等)从图像中提取特征向量。

在这里插入图片描述

image_word_data = pd.read_csv('附件1/ImageWordData.csv')
#图像预处理函数
def preprocess_images(image_folder, image_ids, target_size=(224, 224)):
processed_images = {}
for image_id in image_ids:
image_path = os.path.join(image_folder, image_id)
try:
# 打开图像文件
with Image.open(image_path) as img:
# 调整图像尺寸
img = img.resize(target_size)
# 将图像转换为数组
img_array = np.array(img)


# 对图像数组进行归一化
img_array = img_array / 255.0
processed_images[image_id] = img_array
except IOError as e:
print(f"无法打开或找到图像 {image_path}。错误信息: {e}")
processed_images[image_id] = None
return processed_images
#假设图像位于"附件1/ImageData"文件夹中
image_folder_path = '附件1/ImageData'
processed_images = preprocess_images(image_folder_path, image_word_data['image_id'])
#检查处理过的图像数量和某个示例图像数组的形状
print(f"处理过的图像数量: {len(processed_images)}")
if processed_images:
example_image = list(processed_images.values())[0]
if example_image is not None:
print(f"示例图像数组形状: {example_image.shape}")
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31

2.多模态特征融合

由于文本和图像特征位于不同的特征空间,我们需要采取方法将它们映射到同一个空间,以便进行相似度比较。这可以通过以下方法之一实现:

联合嵌入空间:通过训练一个深度学习模型来同时学习文本和图像的嵌入,使得相似的图像和文本对靠近。

交叉模态匹配网络:设计一个网络,它可以接受一种模态的输入,并预测另一种模态的特征表示。

文本特征提取

from sklearn.feature_extraction.text import TfidfVectorizer
#初始化TF-IDF向量化器
vectorizer = TfidfVectorizer(max_features=1000) # 使用最多1000个词语的词汇量
#将文本数据转换为TF-IDF特征矩阵
tfidf_matrix = vectorizer.fit_transform(preprocessed_captions)
#查看TF-IDF特征矩阵的形状
print(tfidf_matrix.shape)
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7

图像特征提取:

import torch
from torchvision import models, transforms
from PIL import Image
import os
#图像预处理函数
def preprocess_image(img_path):
# 读取图像,转换为RGB(如果是灰度图像)
img = Image.open(img_path).convert('RGB')
# 转换图像
img_t = preprocess(img)
batch_t = torch.unsqueeze(img_t, 0)
return batch_t
#定义预处理流程,确保模型接收三通道的图像
preprocess = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
#你可以在这里选择较小的模型来减少内存使用
#比如使用 ResNet18
model = models.resnet18(pretrained=True)
model.eval() # 设置为评估模式
#修改图像特征提取部分,使用上面定义的preprocess_image函数
def extract_image_features(image_folder, image_ids):
image_features = {}
for image_id in image_ids:
image_path = os.path.join(image_folder, image_id)
try:
batch_t = preprocess_image(image_path)
#batch_t = batch_t.to(device)
with torch.no_grad():
features = model(batch_t)
image_features[image_id] = features.cpu().numpy().flatten()
except Exception as e:
print(f"无法处理图像 {image_path}: {e}")
image_features[image_id] = None
return image_features
#假设图像位于"附件1/ImageData"文件夹中
image_folder_path = '附件1/ImageData'
#调用函数提取特征
image_features = extract_image_features(image_folder_path, image_word_data['image_id'])
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42

特征融合:

#转换图像特征字典为矩阵
image_features_matrix = np.array([features for features in image_features.values() if features is not None])
#特征融合
#这里我们简单地将归一化的图像特征和TF-IDF特征进行连接
#确保TF-IDF特征矩阵是稠密的
tfidf_features_dense = tfidf_matrix.todense()
multimodal_features = np.concatenate((image_features_matrix, tfidf_features_dense), axis=1)
#现在 multimodal_features 矩阵包含了每个样本的融合特征
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8

3.图文检索

根据训练好的模型进行图文检索匹配

检索和排序:根据计算出的相似度,对数据库中的图像进行排序,选出相似度最高的前五张图像。

结果展示:

在这里插入图片描述

在这里插入图片描述

步骤三:基于文本检索图像

与步骤三类似,这里直接展示结果。

在这里插入图片描述

下面内容打开内含详细的视频教学,手把手保姆级,模型高精度,结果有保障!

【腾讯文档】2024泰迪杯数据挖掘助攻合集

docs.qq.com/doc/DVVlhb2xmbUFEQUJL
  • 1
  • 2
  • 3
声明:本文内容由网友自发贡献,不代表【wpsshop博客】立场,版权归原作者所有,本站不承担相应法律责任。如您发现有侵权的内容,请联系我们。转载请注明出处:https://www.wpsshop.cn/w/Monodyee/article/detail/475459
推荐阅读
相关标签
  

闽ICP备14008679号