摘要
针对单模态电力设备热缺陷检测算法在昼夜、恶劣气候或隐蔽场合情况下导致目标信息缺失、漏检等检测效果欠佳的问题,提出一种基于多尺度融合的轻量化目标检测算法AML-YOLO。首先,特征提取采用GhostNet网络实现模型轻量化;其次,形成针对中小目标检测的第四检测层,并在特征融合网络中添加多尺度特征融合路径,加强模型对多模态图像中中小尺寸电力设备特征识别能力。最后,根据模型测试结果得到融合所需参数,进而实现基于检测结果的决策级快速目标检测。自建数据集并进行试验,结果表明与单模态检测算法相比,所提算法的保证推理速度的同时检测精度提升2.1%,验证了所提方法的有效性。
Aiming at the problems that the single-modal thermal defect detection algorithm of power equipment leads to the lack of target information and missed detection in the case of day and night,bad climate or hidden occasions,a lightweight object detection algorithm AML-YOLO based on multi-scale fusion is proposed.First,the feature extraction uses GhostNet network to realize the model lightweight;second,a fourth detection layer for small and medium-sized object detection is formed,and a multi-scale feature fusion path is added to the feature fusion network to strengthen the model's ability to recognize the features of small and medium-sized power equipment in multi-modal images.Finally,the parameters required for fusion are obtained according to the model test results,and then the decision-level fast object detection based on the detection results is realized.The self-built dataset and the test results show that compared with the single-mode detection algorithm,the proposed algorithm can ensure the inference speed and improve the detection accuracy by 2.1%,which verifies the effectiveness of the proposed method.
作者
王鑫华
景超
王慧民
张兴忠
WANG Xin-hua;JING Chao;WANG Hui-min;ZHANG Xingz-hong(Shanxi Energy Internet Research Institute,Taiyuan Shanxi 030000,China;School of Software,Taiyuan University of Technology,Jinzhong,Shanxi 030600,China;School of Artificial Intelligence,Xi'an Jiaotong University,Xi'an Shaanxi 710000,China)
出处
《计算机仿真》
2025年第4期106-113,共8页
Computer Simulation
基金
山西省重点研发计划项目(2022ZDYF100)。
关键词
多模态融合
决策级融合
深度学习
热缺陷诊断
Multimodal fusion
Decision level
Deep learning
Thermal defect diagnosis
