低冗余特征的轻量级物流包裹检测模型

Lightweight Logistics Package Detection Model with Low Redundancy Features

鉴于目前物流包裹目标检测模型在设备资源、检测精度和速度方面存在限制问题,提出了一种名为GSYOLO的低冗余特征的轻量级目标检测网络模型,用于快速检测物流包裹类型。提出了称为GSBlock的轻量级特征提取模块作为骨干网络,从输入图像中提取代表性特征,在保持高精度的同时,对模型进行了相应的瘦身处理,加入多种轻量化模块和无参数注意力机制,显著减少了骨干网的参数和浮点运算量,从而实现了快速推理和低功耗。通过使用自建物流包裹数据集进行对比实验,结果显示,与先进的检测模型YOLOv8s相比,GSYOLO模型的平均精度(mAP)达到了98.6%,模型参数减少了94.75%,FLOPs减少了96%。GSYOLO模型参数和FLOPs显著减少,同时检测精度更高,尤其适用于计算资源受限的物流包裹检测场景。

Given the current limitations in terms of computational resources,detection accuracy,and speed of existing logistic parcel object detection models,a lightweight model called GSYOLO with low redundant features is proposed,which is used for fast detection of logistic parcels.Initially,it proposes a lightweight feature extraction module called GSBlock as the backbone network,which extracts representative features from input images while ensuring high accuracy.The model is further optimized by incorporating various lightweight modules and parameter-free attention mechanisms to significantly reduce the parameters and floating-point operations of the backbone network,thereby achieving fast inference and low power consumption.Comparative experiments using a self-built dataset of logistic parcels demonstrate that the GSYOLO model achieves a mAP of 98.6%,the model parameters have been reduced by 94.75%,with a reduction of in model parameters and a decrease 96%in FLOPs.The GSYOLO model significantly reduces parameters and FLOPs while achieving a higher detection accuracy,making it particularly suitable for logistic parcel detection scenarios with limited computing resources.