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

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，无法有效整合多组图像行为特征，造成识别结果误差偏大，不能进行危险行为即时预警。为此，提出了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方法。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，提取异步交互聚合网络下JDE行为特征，进行人员异常行为角度特征识别，为相邻识别特征交互区域类型输出分配不同加权系数量，输出异步交互聚合识别结果。实验结果表明：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较高，适应性优化效果明显。

Identification of Abnormal Behavior of Personnel in Dangerous Working Area of Ciqarette Factorty Based on Asynchronous Interactive Aggregation Network

Affected by the scattered operation area of the cigarette factory, the large number of operators, and the high similarity of behavior characteristics, it is unable to effectively integrate multiple groups of image behavior characteristics, resulting in a large error in the recognition results, and unable to conduct immediate warning of dangerous behaviors. Therefore, an image recognition method of abnormal behavior of personnel in the dangerous operation area of the cigarette factory based on asynchronous interactive aggregation network is proposed. According to the characteristics of the dangerous operation area of the cigarette factory, combined with the effect of image analysis of personnel abnormal behavior, the JDE behavior characteristics under the asynchronous interactive aggregation network are extracted, the angle characteristics of personnel abnormal behavior are identified, the number of different weighting factors are allocated for the output of the adjacent recognition feature interaction area type, and the asynchronous interactive aggregation recognition results are output. The experimental results show that under the optimization of asynchronous interactive aggregation network, the recognition accuracy of personnel abnormal behavior is significantly improved, the overall recognition effect is stable, and the adaptive optimization effect is obvious.