














摘" 要: 針對現有方法對無人機高速公路護欄損壞檢測存在邊緣信息提取效果差、識別精度低的問題,提出一種基于深度學習的變化檢測模型DDE?BIT。首先,采用深度可分離卷積優化主干網絡Resnet18,減少模型的參數數量,降低計算成本;然后,在主干網絡輸出部分引入ECA注意力模塊,在僅增加少量參數的情況下提高模型的跨通道信息捕捉能力;最后,通過跳躍連接方式對BIT雙時空圖像轉換器的輸出特征進行堆疊,提高模型的上下文信息理解能力。以采集的無人機高速公路護欄損壞圖像為實驗數據,實驗結果表明:DDE?BIT模型的交并比和F1分數分別為90.99%、95.28%,相較于原始模型分別提高了2.71%、1.51%,能夠有效地提取護欄損壞的邊緣信息。
關鍵詞: 護欄損壞檢測; 無人機; ECA注意力機制; 深度可分離卷積; 圖像處理; 信息提取
中圖分類號: TN911.73?34" " " " " " " " " " " " "文獻標識碼: A" " " " " " " " " " " "文章編號: 1004?373X(2025)04?0123?07
DDE?BIT?based UAV highway guardrail damage detection
WANG Yang1, 2, GUO Dudu2, 3, SHUAI Hongbo1, 2
(1. College of Intelligent Manufacturing and Modern Industry, Xinjiang University, Urumqi 830017, China;
2. Xinjiang Key Laboratory of Green Construction and Maintenance of Transport Infrastructure and Intelligent Traffic Control, Xinjiang University,
Urumqi 830017, China; 3. School of Traffic and Transportation Engineering, Xinjiang University, Urumqi 830017, China)
Abstract: In allusion to the problems of poor edge information extraction and low recognition accuracy of existing methods for UAV highway guardrail damage detection, a deep learning based change detection model, DDE?BIT, is proposed. A depth?separable convolutional optimization backbone network, Resnet18, is used to reduce the number of parameters of the model and lower the computational cost. An ECA attention module is introduced into the output part of the backbone network to improve the cross?channel information capturing ability of the model with only a small increase in parameters. The output features of the BIT dual spatio?temporal image converter are stacked by jumping connection to improve the understanding ability of the model context information. Taking the collected UAV highway guardrail damage images as the experimental data, the experimental results show that the intersection ratio and F1 score of the DDE?BIT model are 90.99% and 95.28%, respectively, which are 2.71% and 1.51% higher than that of the original model, and can effectively extract the damaged edge information of the guardrail.
Keywords: guardrail damage detection; UAV; ECA attention mechanism; depth?separable convolution; image processing; information extraction
0" 引" 言
高速公路護欄對保障行車安全、增強交通秩序具有重要意義,是我國道路基礎設施巡檢與養護的重要內容[1?2]。現在的高速公路護欄損壞檢測主要依靠人工進行,存在人力成本高、工作效率低的問題。因此,高效、精確的護欄損壞檢測是目前的研究熱點。現有的護欄損壞自動檢測方法主要是依據路側設備或傳感器實現,如:文獻[3]提出了一種基于無線傳感器的護欄碰撞檢測系統,通過物聯網與路側傳感設備的相互連接實現對護欄的檢測;文獻[4]提出了一種基于數據驅動的護欄故障診斷方法,通過布設監測點收集護欄數據,使用仿真建模的方法實現了對故障護欄的檢測。……