智能检测活体动物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、多物种、常见的活体动物的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常见活体动物的算法模型。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含活体动物。1、数据来源：原始数据使用自研X光安检设备，多角度、多场景下透射多物种、多包装方式的常见活体动物，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，构建形成一个包含活体动物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活体动物特征，通过循证规则来完成活体动物的智能识别，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活体动物的智能检测模型。
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。

An enterprise independently collects X-ray security inspection images of common living animals under multiple environments, angles, forms and species, conducts cleaning, annotation and other processing, and uses these as samples to train and develop an intelligent algorithm model for detecting common living animals. This model can be applied to various security inspection scenarios to accurately and quickly detect whether living animals are contained in the inspected objects.
1. Data Source: The original data was collected using self-developed X-ray security inspection equipment, which transmits images of common living animals of multiple species and packaging methods under multiple angles and scenarios, to establish an original X-ray data image library.
2. Data Processing: Preprocessing including geometric transformation, pixel transformation, denoising, matting and other steps is performed on the collected raw data; the data is annotated using semi-automatic annotation tools to obtain pseudo labels, which are then manually corrected, and an audit mechanism is set up to ensure the accuracy and consistency of annotations, thus constructing a dataset containing X-ray security inspection data of living animals.
3. Training and Generation Rules of Detection Model: The processed and annotated dataset is imported into visual detection algorithm models (e.g., Faster R-CNN model) as sample data for deep learning. The model learns to recognize the features of living animals in the dataset through supervised learning, completes intelligent recognition of living animals via evidence-based rules, and outputs relevant attributes including target category and target location. Furthermore, image attribute information of the detected target objects can be exported, such as image type, image format and acquisition time, etc. The finally generated model is an intelligent detection model capable of accurately identifying living animals.
4. Data Tuning: Hyperparameter tuning is adopted to optimize the model, specifically including learning rate, model structure and size, objective loss function and other aspects, so as to continuously improve the model's detection performance.

搜集汇总

数据集介绍

特点

该数据集包含1480条活体动物的X光安检图像数据，用于训练智能检测算法模型，适用于安检场景中的活体动物检测。数据经过清洗和标注处理，更新频次为半年度。