摘要
针对破损路沿石在街景影像中受到目标多尺度、相似地物干扰以及遮挡等问题,提出了一种面向城市街道两侧破损路沿石检测的CDD-YOLO(convolutional swin transformer deformable decouple-YOLO,CDD-YOLO)模型。依据破损路沿石呈现形状尺度多样性特点,嵌入C3_STR(convolutional swin transformer,C3_STR)模块进行特征融合,增强模型对多尺度特征的感知性能;对于相似地物干扰导致的误检现象,加入可变形卷积模块,利用目标区域自适应特性,提升模型对相似地物的判别能力;为避免因遮挡引起的定位不准确问题,引入解耦检测头结构,增强模型对模糊边界特征的提取能力。在自制的街景破损路沿石数据集上进行验证,分析表明,该方法的precision、recall、F1、IoU 4项评价指标分别达到了82.45%、81.22%、81.01%和80.23%,显著优于其他主流目标检测方法,验证了该方法的有效性和可行性。
Aiming at the problems of multi-scale,similar feature interference and occlusion of broken curb stones in street view images,a CDD-YOLO(convolutional swin transformer deformable decouple-YOLO)model for the detection of broken curb stones on both sides of the urban street is proposed.According to the characteristics of shape and scale diversity of broken curb stones,the features are fused by embedding C3_STR(convolutional swin transformer)module to enhance the perception performance of the model on multi-scale features.For the false detection phenomenon caused by the interference of similar ground objects,by adding the deformable convolution module,the adaptive characteristics of the target region are utilized to improve the discrimination ability of the model to similar ground objects.In order to avoid the problem of inaccurate positioning caused by occlusion,the decoupled detection head structure is introduced to enhance the extraction ability of the model for fuzzy boundary features.It is verified on the self-made data set of broken curb stones in street view,and through the analysis of results,four evaluation indicators,which are precision,recall,F1 and IoU of the method reaches 82.45%,81.22%,81.01%and 80.23%,respectively.The proposed method is significantly superior to other mainstream target detection methods,which verifies its effectiveness and feasibility.
作者
戴激光
李岩
DAI Jiguang;LI Yan(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China)
出处
《遥感信息》
CSCD
北大核心
2024年第3期15-21,共7页
Remote Sensing Information
基金
国家自然科学基金(42071428、42071343)。
关键词
破损路沿石检测
街景影像
目标多尺度
特征融合
解耦检测头
broken curb stone detection
street view imagery
multi-scale target
feature fusion
decoupled detection head
