智能检测无人机算法模型的图像训练数据

企业自主采集多环境、多角度、多形态构造的无人机X光安检图像，进行清洗、标注等处理，并以此为样本训练生成智能检测无人机的算法模型。该模型可应用在各类安检场景中，精准、快速检测被检物中是否包无人机物品。1、数据来源：原始数据使用自研X光安检设备，多角度、多场景下透射各形态的无人机采集并建立其原始的X光数据图例库。
2、数据处理：对收集到的原始数据进行进行包括几何变换、像素变换、去噪、抠图等预处理；并对数据利用半自动标注工具标注得到伪标签，然后使用人工修正标注，并设置审核机制，保证标注的准确性和一致性，构建形成一个包含无人机X光安检数据的数据集。
3、检测模型训练生成规则：将处理及标注好的数据集作为深度学习的样本数据导入视觉检测算法模型（如：FasterRCNN模型），通过监督学习的方式让模型学习识别数据集中无人机特征，通过循证规则来完成无人机的智能识别，并输出相关属性，包括目标品项、目标位置。进一步的还可将被检目标对象的图像属性信息导出，如图像类型、图像格式以及采集时间等，最终生成的模型为可精准识别无人机的智能检测模型。
4、数据调优：选择超参数调优的方式对模型优化，具体的包括学习率、模型结构和尺寸、目标损失函数等，持续提升模型检测性能。

This dataset consists of drone X-ray security inspection images independently collected by enterprises under multiple environments, angles and morphological configurations. After undergoing cleaning, annotation and other preprocessing, these images are used as samples to train an intelligent algorithm model for drone detection. The model can be deployed in various security inspection scenarios to accurately and rapidly detect whether drone-related items are present in inspected objects.

1. Data Source: Raw data is collected using self-developed X-ray security inspection equipment, by capturing transmission images of drones in various forms under multiple angles and scenarios, to establish a raw X-ray image dataset for drones.

2. Data Processing: Preprocessing operations including geometric transformation, pixel transformation, denoising and matting are conducted on the collected raw data. Subsequently, semi-automatic annotation tools are utilized to generate pseudo-labels, followed by manual correction of annotations and establishment of an audit mechanism to ensure the accuracy and consistency of annotations, thereby constructing a dataset containing drone X-ray security inspection data.

3. Training and Generation Rules of the Detection Model: The processed and annotated dataset is imported into a visual detection algorithm model (e.g., Faster R-CNN model) as deep learning training samples. The model learns to recognize drone features in the dataset via supervised learning, completes intelligent drone recognition through evidence-based rules, and outputs relevant attributes including target item type and target location. Additionally, image attribute information of the detected target objects can be exported, such as image type, image format and acquisition time. The final generated model is an intelligent detection model capable of accurately identifying drones.

4. Data Tuning: Hyperparameter tuning methods are adopted to optimize the model, specifically including learning rate, model structure and size, target loss function and others, to continuously enhance the model's detection performance.

搜集汇总

数据集介绍

特点

该数据集是由浙江啄云智能科技有限公司提供的无人机X光安检图像训练数据，包含981条记录，用于训练智能检测无人机的算法模型，适用于安检场景中的无人机检测。