现代电子技术2025，Vol.48Issue(11)：51-56,6.DOI:10.16652/j.issn.1004-373x.2025.11.008
基于改进Mask RCNN的金刚石微粉分割算法
Diamond micro powder segmentation algorithm based on improved Mask RCNN
摘要
Abstract
A diamond micro powder segmentation algorithm based on improved Mask RCNN(region-based convolutional neural network)is proposed to cope with the segmentation difficulties caused by the complex and varied shapes,high color similarity,and mutual adhesion of diamond micro powder.The residual structure is redesigned based on the original backbone network and an efficient channel attention mechanism is introduced to enhance the model's ability to extract complex features without increasing network parameters.On the basis of the actual size and the irregularity of the shape of the diamond micro powder,the Anchor box is optimized to enhance the detection box fit and further improve the image segmentation accuracy of the model.The DIoU-NMS module is used to improve the screening means of detection boxes,so as to avoid the missed detection of adhered diamond micro powder.The experimental results show that on the self-built diamond micro powder dataset,the mean average precision(mAP)of the improved Mask RCNN algorithm for segmentation of diamond micro powder is 75.51%,which is 2.86%higher than that of the standard Mask RCNN algorithm.The segmentation accuracy of the proposed algorithm has been improved significantly,so the proposed algorithm is feasible for accurate segmentation of diamond micro powder.关键词
深度学习/实例分割/金刚石微粉/注意力机制/Mask RCNN/DIoU-NMSKey words
deep learning/instance segmentation/diamond micro powder/attention mechanism/Mask RCNN/DIoU-NMS分类
信息技术与安全科学引用本文复制引用
李文开,王莉,牛群峰,王涛..基于改进Mask RCNN的金刚石微粉分割算法[J].现代电子技术,2025,48(11):51-56,6.基金项目
河南省科技攻关项目(242102220094) （242102220094）
河南工业大学创新基金支持计划专项(2022ZKCJ03) （2022ZKCJ03）
