Region segmentation algorithm for mine conveyor belt based on transfer learning and feature attention enhanced PP_LiteSeg
WANG Yuanbin
CHANG Wenjian
CHEN Xiaojing
WANG Xu
LIU Jia
Abstract：To achieve accurate detection of unauthorized interactions in mine conveyor belt areas,an improved PP_LiteSeg semantic segmentation algorithm integrating transfer learning and a fea-ture attention enhancement mechanism was proposed to tackle the challenges of sample scarcity,multi-scale targets,and blurred edges in underground images.A Feature Attention Pyramid Pool-ing module was designed to fuse multi-scale features,thereby enhancing the perception of target de-tails across various scales.A lightweight multi-level unified attention fusion decoder network was constructed to strengthen the discriminative ability between targets and complex backgrounds.Ad-ditionally,a transfer learning strategy using pre-trained weights was introduced to accelerate conver-gence on the conveyor belt dataset.The results show that the proposed algorithm achieves an accu-racy of 99.2%,a mean intersection over union of 90.4%,a Dice similarity coefficient of 94.8%,and a kappa coefficient of 95.4%.These metrics show significant improvements compared with mainstream networks,including DeepLabv3,U-Net,EfficientNet,ShuffleNet,MobileNetV3,and the original PP_LiteSeg.This study provides high-precision technical support for unauthorized behavior detection and intelligent safety supervision in mine conveyor belt zones.
Keywords：semantic segmentationPP_LiteSegfeature attention pyramid poolingLMUAFDtransfer learning
Publication Date：2026-03-31
Online Publishing Date：2026-04-08(First online date of this platform, not the publication date of the document)
Pages：13( 435-446,545 )