摘要
随着城市化步伐的不断加快,电动车已成为城市居民日常出行的重要工具,许多居民为了充电方便,选择将电动车搬进居民楼,并通过电梯运输,这一行为导致了居民楼内电动车安全事故频发.因此,开发一种高效且准确的电梯内电动车检测算法显得尤为重要.首先,使用 FasterNet 中的 FasterNet Block替换C2f中的Bottleneck,其降低了模型的计算量,提升了检测速度.在YOLOv8原结构中引入SEAttention注意力机制,把重要的特征进行强化来提升准确率.替换损失函数为Inner-CIoU损失函数,以提升模型检测性能和泛化能力.经过实验验证,改进后的YOLO-FSI模型在电梯内电动车数据集上的mAP50为93.7％,相较于原模型,参数量减少了23.3％,检测速度更是提升了 5.2帧/秒.综上所述,YOLO-FSI模型可以有效提升电梯内电动车的检测能力,并且做到了轻量化及快速推理.
Abstract
With the accelerating pace of urbanization,electric vehicles have become an important tool for urban residents to travel daily.Many residents choose to move electric vehicles into resi-dential buildings and transport them through elevators for the convenience of charging.This be-havior has led to frequent safety accidents of electric vehicles in residential buildings.Therefore,it is particularly important to develop an efficient and accurate detection algorithm for electric vehicles in elevators.Firstly,the FasterNct Block in FastcrNct is used to replace the Bottleneck in C2f,which reduces the computational complexity of the model and improves the detection speed.The SEAttcntion attention mechanism is introduced into the original structure of YOLOv8 to strengthen the important features to improve the accuracy.The loss function is re-placed by the Inner-CIoU loss function to improve the detection performance and generalization ability of the model.After experimental verification,the mAP50 of the improved YOLO-FSI model on the electric vehicle data set in the elevator is 93.7％.Compared with the original model,the parameter amount is reduced by 23.3％,and the detection speed is increased by 5.2 frames/second.In summary,the YOLO-FSI model can effectively improve the detection abili-ty of electric vehicles in elevators,and achieve lightweight and fast reasoning.
NETL
NSTL
万方数据