核电厂焊缝射线质量智能评定模型数据

1.对核电厂焊缝射线质量、缺陷进行智能精准识别与评定，为核电厂焊缝射线质量智能评定模型提供训练数据，提高核电厂焊缝射线评定标准的准确性，进一步提升核电厂焊缝射线质量。
2.能够为核电厂焊缝质量控制提供决策依据，通过焊缝质量、缺陷训练数据持续优化模型检测标准，进一步保障核电站设备安全和长期稳定运行。数据收集和处理：
步骤1数据收集：方式一：通过使用数字射线检测设备采集获取原始焊缝图像数据；方式二：通过高分辨率扫描设备对传统射线底片影像数据转化为原始焊缝图像数据。
步骤2数据处理：对收集到的原始焊缝图像数据进行图像预处理，去除图像噪声、矫正图像畸变，统一图像尺寸（将图像的高度（imageHeight）和宽度（imageWidth）统一为3580×780（像素），确保原始焊缝图像数据质量和一致性。
步骤3数据标注：使用图像标注工具对原始焊缝图像数据的Filename(文件名)、Shapes（形状）、imagePath(路径)、imageData（图像数据）、imageHeight（图像高度）、imageWidth（图像宽度）等字段进行数据标注。其中：Filename(文件名)按照image_XXX.json按序号依次标注；Shapes（形状）字段中的label（标签）用于标注缺陷类型，缺陷类型标签包括5个缺陷等级：liewen（裂纹）、qikong（气孔）、jiazha(夹渣)、weironghe（未熔合）、weihantou(未焊透)，points(位置)用于标记缺陷在图像中的位置，通过坐标的形式进行标注，shape_type（形状类型）为用于框选缺陷的形状，在本数据标注过程中均使用rectangle（矩形）框进行框选，在同一个焊缝图像数据中可能存在多个缺陷，在进行数据标注的过程中Shapes（形状）可能包含多个缺陷；imagePath(路径)用于标记生成标注文件的存储位置，存储在图像的同一路径（文件夹）下；imageData（图像数据）为对通过数据收集阶段获取到的原始焊缝图像数据进行标注，按照image_XXX.tiff进行命名存储；对图像高度（imageHeight）和宽度（imageWidth）分别标注为3580和780。通过以上数据标注过程确保为后续模型训练提供高质量的标注数据。
通过使用核电厂焊缝射线质量智能评定模型数据能够构建例如：卷积神经网络（CNN）的自动识别模型，实现核电厂焊缝射线图像缺陷的自动识别并和缺陷类型的标注，辅助人工评定工作，提升检测效率和准确度。

1. This dataset is used for intelligent and accurate identification and evaluation of radiographic quality and defects of nuclear power plant welds, providing training data for the intelligent evaluation model of nuclear power plant weld radiographic quality, improving the accuracy of weld radiographic evaluation standards for nuclear power plants, and further enhancing the radiographic quality of nuclear power plant welds.
2. It can provide decision-making basis for nuclear power plant weld quality control, continuously optimize the model detection standards through weld quality and defect training data, and further ensure the safe and long-term stable operation of nuclear power plant equipment.

Data Collection and Processing:
Step 1 Data Collection: Method 1: Collect original weld image data using digital radiographic detection equipment; Method 2: Convert traditional radiographic film image data into original weld image data via high-resolution scanning equipment.
Step 2 Data Processing: Perform image preprocessing on the collected original weld image data, including removing image noise, correcting image distortion, and unifying the image dimensions (unifying imageHeight and imageWidth to 3580 × 780 pixels) to ensure the quality and consistency of the original weld image data.
Step 3 Data Annotation: Use image annotation tools to annotate fields such as Filename, Shapes, imagePath, imageData, imageHeight, and imageWidth of the original weld image data. Specifically:
- Filename: Annotate sequentially in the format of "image_XXX.json" according to the serial number;
- The label in the Shapes field is used to annotate defect types, which include 5 defect levels: liewen (crack), qikong (blowhole), jiazha (slag inclusion), weironghe (incomplete fusion), weihantou (incomplete penetration). The points field is used to mark the position of defects in the image through coordinate annotation. The shape_type is the shape used to frame the defects, and rectangle frames are used for all annotations in this dataset. Multiple defects may exist in the same weld image, so the Shapes field may contain multiple defect entries;
- imagePath is used to mark the storage location of the generated annotation files, which are stored in the same folder as the corresponding images;
- imageData corresponds to the original weld image data acquired in the data collection stage, which is named and stored in the format of "image_XXX.tiff";
- The imageHeight and imageWidth are annotated as 3580 and 780 respectively.
The above data annotation process ensures that high-quality annotated data is provided for subsequent model training.

By using this nuclear power plant weld radiographic quality intelligent evaluation model dataset, automatic recognition models such as Convolutional Neural Networks (CNN) can be constructed to realize automatic identification of defects in nuclear power plant weld radiographic images and annotation of defect types, assisting manual evaluation work, and improving detection efficiency and accuracy.

搜集汇总

背景与挑战

背景概述

该数据集包含1001条核电厂焊缝射线图像数据，用于智能评定模型训练，标注了裂纹、气孔等5种缺陷类型，旨在提升焊缝质量检测的准确性和核电站设备安全。数据以CSV格式存储，每年更新一次，支持卷积神经网络等模型开发。