布匹质检专利图搜图AI模型训练数据

缺陷检测在纺织领域具有繁多的发明专利申请，此类专利申请的摘要、题目无法准确表示缺陷的类型，为了检索相似布匹缺陷的公开方法，使用者需要大量尝试检索式，导致效率低下，且不准确。
知产行业正在数智化转型，通过设计和构建数据集训练AI模型，可以让用户用布匹缺陷图像搜索相关专利，极大提升用户体验和检索效率。

该数据与专利之间的联系在于能够完成以下两个应用：
1、以图搜图:用于交叉比对专利申请中图像的重复等异常情况,也可以通过相似图像(例如牛仔裤撕裂的特征)搜索相关专利文件,用于快速检索。
2、以特征搜图:包围框标注的特征可以训练目标检测模型,检索时可以按照缺陷类型来检索相关专利。
对非特定领域也有很强的现实意义,可以面向常见的布料缺陷训练缺陷检测AI模型。

数据集包含191张高质量的布匹纹理图片以及对应每个图片的边界框和缺陷类别详细标注信息。

为了上述两个专利专用应用,数据集去除了表征单一或质量差的图片,仅保留表征丰富且缺陷表现清晰的高质量图片。

Defect detection has seen a high volume of invention patent applications in the textile industry. Yet the abstracts and titles of these applications fail to accurately specify the types of defects. To retrieve public methods for similar fabric defect cases, users have to test numerous search queries, leading to low efficiency and unsatisfactory accuracy.
The intellectual property (IP) industry is currently undergoing digital and intelligent transformation. By designing and building datasets to train AI models, users can search for relevant patents using fabric defect images, which significantly enhances user experience and retrieval efficiency.
The dataset supports two core patent-oriented applications as follows:
1. Image-based Search: Used for cross-checking anomalies such as duplicate images in patent applications, and also for rapid retrieval of relevant patent documents by searching similar images (e.g., features of torn jeans).
2. Feature-based Search: The bounding box-annotated features can be used to train object detection models, allowing users to retrieve relevant patents based on defect types during search.
This dataset also holds strong practical value for non-specialized fields, as it can be utilized to train defect detection AI models for common fabric defects.
The dataset consists of 191 high-quality fabric texture images, paired with detailed annotation information including bounding boxes and defect categories for each individual image.
For the two patent-specific applications mentioned above, the dataset has filtered out images with single defect representations or poor quality, only retaining high-quality images with rich feature representations and clearly visible defect manifestations.

搜集汇总

数据集介绍

特点

布匹质检专利图搜图AI模型训练数据包含191张高质量布匹纹理图片及详细标注，用于训练AI模型以提升纺织领域专利检索效率和准确性。