智能检测香烟算法模型的训练数据

本数据集包含多角度、多场景下的香烟的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智能检测香烟的算法模型，实现对藏匿在其他物品中或伪装成其他物品等复杂环境下的香烟的精准识别。1、数据来源：应用X射线光源多角度、多场景下透射香烟物品，采集并建立其原始的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，并对采集的香烟物品的X光图像进行抠图处理。将抠出的图像与多场景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：将处理后的香烟的X光安检图像和场景X光图像通过密度统计（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、精准识别香烟的智能检测模型。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5，ρ为密度值，α、β指系数）检测模型可对多场景下香烟的精准识别，同时将目标物的位置及所在X光图像信息记录标出。进一步的还可根据目标物位置信息推算目标物尺寸信息。

This dataset contains X-ray security inspection images of cigarettes captured from multi-angle and multi-scenario perspectives. Through processing steps such as annotation, matting, data augmentation and image fusion, it can serve as high-quality samples to train intelligent cigarette detection algorithm models, enabling accurate identification of cigarettes hidden among other items or disguised as other items in complex environments.
1. Data Source: Apply X-ray light sources to transmit cigarette items from multi-angle and multi-scenario perspectives, collect and establish an original X-ray image library for cigarette items.
2. In-depth Data Processing: Pre-annotate the coordinate positions and item categories of the collected original X-ray images, and perform matting processing on the X-ray images of the collected cigarette items. Perform augmentation processing including geometric transformation and pixel transformation on the matting-extracted images and multi-scenario images respectively.
3. Algorithm Rules for Generating Detection Models: Merge the processed X-ray security inspection images of cigarettes and scene X-ray images according to the region matching principle based on density statistics (pixel values represent physical density values). The fused region mask is used as the data label, and the fused image is used as the deep learning sample data. By adjusting the position of the matting-extracted region within the scene image region, different average density differences can be obtained to train and generate an intelligent detection model capable of accurately locating and identifying cigarettes. The region matching principle follows the formula: Mask*(α*ρ_matting_image + β*ρ_scene_image). The post-fusion image processing formula follows: Mask*(α*ρ_matting_image + β*ρ_matting_image) + (1-Mask)*ρ_scene_image. (In the above formulas: Mask is the image mask, with a value of 1 within the target region and 5 outside the target region; ρ represents the density value; α and β are coefficients.)
The detection model can accurately identify cigarettes in various scenarios, and simultaneously record and mark the position of the target object and the information of the X-ray image where it is located. Furthermore, the size information of the target object can also be inferred based on its position information.