【论文简读】 Deep web data extraction based on visual

视觉信息网页分割
提出一种基于卷积神经网络(CNN)的数据区域定位方法,并结合视觉信息进行网页分割。该方法通过13级CNN网络实现区域定位,实验结果表明在特定数据集上取得了较高的精度。
Deep web data extraction based on visual information processing
作者 J Liu 上海海事大学 2017 AIHC会议登载
引用 Liu J, Lin L, Cai Z, et al. Deep web data extraction based on visual information processing[J]. Journal of Ambient Intelligence & Humanized Computing, 2017(1):1-11.

简介

一种基于卷积神经网络(CNN)的数据区域定位方法
结合视觉信息进行网页的分割(作者命名为VIBS)

1、CNN 基于卷积神经网路进行区域定位

CNN网络结构由3个阶段组成,如图所示 。第一阶段设置卷积层和汇集层以学习图像的特征。第二阶段是设置展平图层所必需的,展平图层会将卷积图层和合并图层生成的特征图转换为一维矢量,以计算完整的连接图层。除了为输出设置最后一个完全连接之外,第三阶段设置多个连接层以过滤先前层学习的特征。
clipboard.png
网络架构设计使用13级CNN,包括5个采样层(S),6个卷积层(C)和2个完全连接层。

clipboard.png

据区域检测的标准IOU,如果IOU  > 50%,则数据区域被视为正样本。

clipboard.png

区域定位主要步骤流程图如下

clipboard.png

2、基于视觉信息的网页分割方法 VIBS

clipboard.png

clipboard.png

clipboard.png

3、实验结果和分析

数据集(Lianjia、Complanet、Fangjia)

58,500个样本数据集,其中有195种具有不同大小和不同位置的图像样本,包含数据区域,总共300个组。

实验结果

clipboard.png

clipboard.png

clipboard.png

clipboard.png

clipboard.png

clipboard.png

clipboard.png

clipboard.png

clipboard.png

总结

总体看下来,文章的创新意义大于实际意义吧,这么高的精确度,感觉像是过拟合了,而且速度不可能这么快,应该是把网页先行保存成图片了的,文章写得很不错,对比什么的体系也比较完善,就是有些地方没有讲清楚,比如能否divide的判定等。

### Robot Vision Technology and Applications in Computer Science Robot vision, an integral part of robotics, leverages the principles from both machine learning (ML) and computer vision (CV). The "machine" aspect refers to statistical models while "learning" involves fitting these models using data[^2]. This combination allows robots not only to perceive their environment but also interpret it meaningfully. #### Key Components of Robot Vision Systems A typical robot vision system consists of several key components: - **Image Acquisition**: Cameras or other sensors capture images or video streams. - **Preprocessing**: Raw image data undergoes preprocessing steps such as noise reduction, normalization, etc., preparing them for further analysis. - **Feature Extraction**: Identifies important features within the captured imagery through techniques like edge detection, corner finding, texture analysis, among others. - **Object Recognition & Classification**: Utilizes ML algorithms including deep neural networks trained on large datasets to recognize objects accurately even under varying conditions. - **Pose Estimation**: Determines spatial orientation relative to known reference points aiding navigation tasks significantly. - **Action Planning Based On Visual Input**: Integrates visual information into decision-making processes enabling autonomous behavior execution based upon perceived surroundings. ```python import cv2 from sklearn import svm # Example code snippet demonstrating object recognition with SVM classifier def train_classifier(training_data): clf = svm.SVC() clf.fit(training_data['features'], training_data['labels']) return clf def predict_object(image, model): feature_vector = extract_features(image) prediction = model.predict([feature_vector]) return prediction ``` The development trend towards more sophisticated systems has seen increased reliance on artificial intelligence methodologies especially those rooted deeply within ML paradigms. However, there remains a cautionary note regarding overemphasis on immediate practicality at expense of foundational scientific inquiry which could be detrimental long term progress. --related questions-- 1. What are some common challenges faced when implementing real-time robotic vision? 2. How does pose estimation contribute specifically to improving robot autonomy? 3. Can you provide examples where traditional CV methods outperform modern DL approaches in certain scenarios related to robotics? 4. In what ways can integrating multiple sensor modalities enhance performance beyond single-camera setups used primarily today?
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值