| 引用本文: | 张焕龙,周钶燕,王延峰,等.基于掩码记忆的无人机电力设备分割跟踪方法研究[J].电力系统保护与控制,2024,52(21):140-150.[点击复制] |
| ZHANG Huanlong,ZHOU Keyan,WANG Yanfeng,et al.A segmentation and tracking method for UAV power equipment based on mask memory[J].Power System Protection and Control,2024,52(21):140-150[点击复制] |
|
| 摘要: |
| 无人机巡检是保障新型电力系统安全运行的重要手段，视频目标跟踪技术能够有效提高巡检精度。针对实际电力场景中背景复杂、目标形变大等挑战，提出一种基于掩码记忆的无人机电力设备分割跟踪方法。首先，设计了基于掩码记忆的L2匹配分割方法，使用负平方欧几里得距离来计算帧间亲和度，利用构建的历史掩码信息进行目标匹配，提升设备的粗分割精度。其次，为了减少记忆冗余，提出变化感知的记忆更新机制，通过图像和掩码联合评估目标变化程度，设计变化度置信度阈值，以此决策是否进行记忆更新。最后，将分割技术融入判别式相关滤波的跟踪框架中，实现电力设备鲁棒分割跟踪。该方法分别在通用数据集VOT2018、无人机数据集UAV123和实际电力场景数据集中进行测试。实验结果表明所提方法能有效提升无人机巡检精度，并为电力设备安全运行提供技术参考。 |
| 关键词: 目标跟踪 目标分割 记忆更新 电力设备跟踪 无人机巡检 |
| DOI：10.19783/j.cnki.pspc.240250 |
| 投稿时间：2024-03-05修订日期：2024-08-01 |
| 基金项目:国家自然科学基金项目资助(62272423， 62072416，62006213，62102373)；河南省杰出青年科学基金项目资助(2300421055)；河南省重点研发专项资助(241111210400) |
|
| A segmentation and tracking method for UAV power equipment based on mask memory |
| ZHANG Huanlong1,ZHOU Keyan1,WANG Yanfeng1,TIAN Yangyang2,ZHAI Denghui3,LI Zhe2,XU Dan4,LIU Heng5 |
| (1. College of Electrical and Information Engineering, Zhengzhou University of Light Industry, Zhengzhou 471200, China;
2. State Grid Henan Electric Power Research Institute, Zhengzhou 450052, China; 3. China Electric Equipment Group
Science and Technology Research Institute Co., Ltd., Shanghai 200000, China; 4. XJ Electric Co., Ltd.,
Xuchang 461000, China; 5. Pinggao Group Co., Ltd., Pingdingshan 467001, China) |
| Abstract: |
| UAV inspection is vital for ensuring the safe operation of new power systems, and visual object tracking technology can effectively improve inspection accuracy. Given the challenges of complex background and large target deformation in an actual electrical power scene, a segmentation and tracking method for UAV power equipment based on mask memory is proposed. First, an L2 mask memory matching method improves segmentation accuracy by calculating frame affinities using the negative squared Euclidean distance and uses historical mask information for target matching. Second, to reduce memory redundancy, a change-aware memory updating mechanism is introduced. The degree of target change is jointly assessed by image and mask, and a confidence threshold-based memory update strategy is designed. Finally, the segmentation strategy is combined with the DCF tracking framework to realize robust segmentation tracking of power equipment. The proposed method is evaluated on VOT2018 (generalized dataset), UAV123 (UAV dataset), and a real power scenario dataset. Experimental results demonstrate that the proposed method effectively enhances UAV inspection accuracy and provides valuable technical references for the safe operation of power equipment. |
| Key words: object tracking target segmentation, memory update power equipment tracking UAV inspection |
