智能检测枪支及仿制品模型X光图像训练数据

本数据集包含多角度、多场景下的枪支及仿制品X光安检图像，通过对图像的标注、抠图、增强、融合等处理，可作为优质样本训练生成枪支及仿制品智能检测模型，实现算法模型对枪支及仿制品的精准识别。本数据集枪支及仿制品图例形态丰富、场景多样、更新及时。通过本数据集的深入应用，可有效提升枪支及仿制品智能检验检测模型的检测精度、速度，提升模型性能。1、数据来源：应用X射线光源多维度、多角度、多场景下透射枪支及仿制品，采集并建立原始X光数据图例库。
2、数据深度处理：对采集到的原始X光图像预标注坐标位置和品项类别，并对枪支类物品图像进行抠图处理。将抠出的枪支及仿制品图像与多场景的图像分别进行几何变换、像素变换等增广处理。
3、检测模型生成算法规则：将处理后的枪支类X光图像和场景图像通过密度统计（像素值代表实物密度值）依据区域匹配原则进行融合，融合区域掩模作为数据标签与融合后的图像作为深度学习样本数据。还可通过调整抠图区域在场景图像区域的位置，获得不同的平均密度差值，训练生成可精准定位、精准识别枪支及仿制品的智能检测模型。区域匹配原则按照Mask*(α*ρ抠图图像+β*ρ场景图像)，融合后的图像处理公式按照Mask*(α*ρ抠图图像+β*ρ抠图图像)+（1-Mask）*ρ场景图像。（所述公式中：Mask为图像掩膜，图像目标区域值为1，目标区域外值为0，ρ为密度值，α、β指系数）检测模型可对多场景下的枪支类物品精准识别，并将目标物为枪支的物品检出，同时将目标物的位置及所在X光图像信息记录标出。还可根据目标物位置信息推算目标物尺寸信息。

This dataset comprises X-ray security screening images of firearms and their replicas captured from multiple angles and scenarios. Through processing steps including image annotation, matting, enhancement and fusion, it can serve as high-quality samples to train intelligent detection models for firearms and their replicas, enabling the algorithmic models to accurately identify such items. The firearms and replicas included in this dataset feature diverse forms, varied scenarios and timely updates. Through in-depth application of this dataset, the detection accuracy and speed of intelligent inspection and detection models for firearms and their replicas can be effectively improved, and the overall model performance can be enhanced.
1. Data Source: X-ray transmission data of firearms and their replicas under multi-dimensional, multi-angle and multi-scenario conditions using X-ray light sources are collected to establish an original X-ray image database.
2. Data Deep Processing: Pre-label the coordinate positions and item categories of the collected original X-ray images, and perform matting processing on the images of firearm-related items. Augmentation processing including geometric transformation and pixel transformation is then conducted separately on the matting-extracted images of firearms and their replicas and images of various scenarios.
3. Detection Model Generation Algorithm Rules: The processed firearm X-ray images and scenario images are fused according to the region matching principle via density statistics (where pixel values represent physical density values). The fused region mask is used as the data label, and the fused image serves as the deep learning sample data. By adjusting the position of the matting region within the scenario image region, different average density differences can be obtained to train and generate an intelligent detection model capable of accurately locating and identifying firearms and their replicas. The region matching principle follows the formula Mask*(α*ρ_matting_image + β*ρ_scenario_image), and the post-fusion image processing formula follows Mask*(α*ρ_matting_image + β*ρ_matting_image) + (1-Mask)*ρ_scenario_image. (In the aforementioned formula: Mask is the image mask, with a value of 1 within the target region and 0 outside the target region; ρ denotes the density value; α and β are coefficients.) The detection model can accurately identify firearm-related items across multiple scenarios, detect items classified as firearms, and record and mark the position of the target object and the X-ray image information where it is located. It can also infer the size information of the target object based on its position information.