01683nas a2200265   4500000000100000008004100001653003100042653003400073653002500107653002200132653001800154653002300172653002000195653002300215653001800238653003700256653003400293653002000327653001100347100001600358245016200374856014400536490001000680520072700690                                       d10aConvolution neural network10aConvolutional neural networks10aDeep neural networks10aDetection methods10aDigital image10aDigital protection10aEdge structures10aImage segmentation10aImage texture10aInner convolution neural network10aIntangible cultural heritages10aLong embroidery10aYolov81 aYuanxiao Ba00aResearch on GSEm-Net detection method of embroidery pattern lightweight operator for digital protection of Gansu intangible cultural heritage Long Embroidery  uhttps://www.scopus.com/inward/record.uri?eid=2-s2.0-105001143728&doi=10.1117%2f12.3059274&partnerID=40&md5=215645e0cdb65d1c4fe653b28da0bbe20 v135503 aThe digital image of embroidery has complex edge structure and high repetition rate texture features. Post-processing such as region detection is a challenging task. A long embroidery image data set for learning and training is constructed. Based on the deep network and multi-scale detection structure, the lightweight detection network model GSEm-Net is introduced to detect the long embroidery pattern in real time. Experiments show that compared with yoloV8, the detection accuracy of this method is equivalent, and the detection speed is 1.5 times that of yoloV8, but the generated model is smaller, and it is easier to deploy to low-computing edge devices, which effectively improves the target detection efficiency.