| 基于异步交互聚合网络的卷烟厂危险作业区人员异常行为图像识别 |
投稿时间：2023-02-01 修订日期：2023-04-25  点此下载全文 |
| 引用本文： |
| 摘要点击次数: 35 |
| 全文下载次数: 0 |
|
|
| 中文摘要:受到卷烟厂作业区域分散，作业人员多，行为特征相似度高的影响，传统识别方法无法有效整合多组图像行为特征，造成识别结果误差偏大，无法起到危险行为即时预警的效果。根据卷烟厂危险作业区域特点，结合人员异常行为图像分析效果，引入异步交互聚合网络对其识别行为特征进行优化。优化共计分为3部分：第一部分，人员异常行为角度特征识别；第二部分，异步交互聚合网络下JDE行为特征提取；第三部分；异步交互聚合识别结果输出；通过对提出方法识别效果的数据调试表明：在异步交互聚合网络优化下，人员异常行为识别准确率得到明显提升，整体识别效果稳定性得到进一步提升，方法适应性优化效果明显。 |
| 中文关键词:异步交互聚合网络 人员 异常行为 图像识别 |
| |
| Identification of abnormal behavior of personnel in dangerous working area based on asynchronous interactive aggregation network |
|
|
| Abstract:Influenced by the scattered operating areas of cigarette factories, the large number of operators and the high similarity of behavioral characteristics, the traditional identification methods cannot effectively integrate the behavioral characteristics of multiple groups of images, resulting in a large error of identification results and unable to play the effect of immediate warning of dangerous behavior. According to the characteristics of dangerous operation area in cigarette factory and the image analysis effect of personnel abnormal behavior, the asynchronous interactive aggregation network is introduced to optimize its recognition behavior characteristics. The optimization is divided into three parts: the first part is the identification of abnormal behavior features; the second part is the JDE behavior feature extraction under the async interaction aggregation network; the third part is the output of the async interaction aggregation recognition results; the data debugging of the identification effect of the proposed method shows that: under the optimization, the recognition accuracy of abnormal behavior is significantly improved, the stability of the overall recognition effect is further improved, and the method adaptation optimization effect is obvious. |
| keywords:asynchronous interactive aggregation network personnel abnormal behavior image recognition |
| 查看全文 查看/发表评论 下载pdf阅读器 |
位访问者,今日访问
次