【摘要】 针对X光安检图像违禁品检测模型参数量高、浮点型计算量大、检测速度慢不易实际工业部署等问题，提出一种改进的轻量级违禁品目标检测模型--EGD-YOLOv5。首先在YOLOv5中引入C3Ghost和Ghost模块，减少特征信道融合过程中的浮点运算，提高特征表达性能；然后在特征提取网络中融入ECA注意力机制提取到检测目标更多的特征信息；最后，将IOU＿nms修改为DIOU＿nms来优化损失函数，提高遮挡目标的检测精度。试验结果表明，改进后的算法降低了模型的参数量和浮点型计算量，检测速度较快，易于部署在资源有限的设备中。更多还原

【Abstract】 Aiming at the problems of high number of parameters, large amount of floating-point calculation, slow detection speed and difficult actual industrial deployment of X-ray security inspection image contraband detection model, an improved lightweight contraband target detection model----EGD-YOLOv5 is proposed. Firstly, the C3Ghost and Ghost modules are introduced in YOLOv5 to reduce the floating-point operation in the process of feature channel fusion and improve the feature expression performance. The ECA attention mechanism is integrated into the feature extraction network to extract more feature information from the detection target. Finally, the IOU＿nms is modified to DIOU＿nms to optimize the loss function and improve the detection accuracy of occlusion targets. The experimental results show that the improved algorithm reduces the parameter amount and floating-point calculation of the model, and the detection speed is faster and easy to deploy in the equipment with limited resources.更多还原