Wyniki wyszukiwania - BazTech

Ograniczanie wyników

1 2025

Znaleziono wyników: 1

Liczba wyników na stronie

Wyniki wyszukiwania

Wyszukiwano:
w słowach kluczowych: parcel sorting

Sortuj według:

Ogranicz wyniki do:

Intelligent logistics express parcel real-time detection system based on improved YOLOv8

Gao Kailan, Wang Xuelin, Xu Yulong, Sładkowski Aleksander

Transport Problems

2025

T. 20, z. 1

179--192

In intelligent warehousing and logistics centers, efficient sorting and transportation of express parcels are critical tasks for daily operations. In order to handle the large number of parcels per day, existing inspection and sorting systems face challenges such as high computational costs, insufficient inspection accuracy and poor real-time response capability. These issues restrict the deployment of such systems on embedded devices and impact the overall efficiency of logistics operations. This is particularly urgent in intelligent sorting, where real-time intelligent recognition and efficient sorting transportation are crucial. In order to solve these problems, a lightweight YOLOv8-SCS-CE algorithm based on the YOLOv8n algorithm is proposed in this paper, which can quickly realize express and parcel detection. A lightweight module SCS is also used to improve the Shufflenetv2 network by utilizing the channel attention mechanism and nested structure to overcome the effect of the lightweight architecture on the detection accuracy. First, the SCS module is used to replace the backbone structure of the YOLOv8n network, which improves the Shufflenetv2 network using channel attention mechanisms and nested structures, overcoming the impact of lightweight architectures on detection accuracy. Next, to enhance the network’s feature extraction capability, the ECA attention mechanism is incorporated into the forward network of C2f, forming the C2f_ECA lightweight feature extraction module, which effectively solves the issue of extracting small defect features in complex backgrounds. Experimental results show that the YOLOv8-SCS-CE model reduces the number of parameters by 59.2%, decreases the computational complexity by 54.3%, and the weight file size is only 2.8MB, a reduction of 55.5% compared to the original files. The mAP@50 of the model is only 0.8% lower than that of the original mode. The lightweight network constructed in this study recognizes fast and few parameters, which provides a key technology for the courier parcel detection industry.