基于麻雀搜索算法改进的YOLOv7-ECA-SSA模型的车辆检测
Vehicle detection based on improved YOLOv7-ECA-SSA model with sparrow search algorithm
陈红 1张乐1
作者信息
- 1. 西安工业大学电子信息工程学院 西安 710021
- 折叠
摘要
为解决复杂背景下小目标车辆检测存在的误检、漏检等现象,创新性提出一种改进YOLOv7网络的目标检测算法.首先,为解决小目标车辆存在次要信息干扰问题,将高效通道注意力(ECA)机制融于YOLOv7模型的主干网络特征层,通过自适应学习来增强目标区域信息权重占比,抑制无关信息;其次,为解决神经网络检测模型训练的超参数随机经验设定性问题,将麻雀搜索算法(SSA)对检测模型训练超参数进行优化,通过内外双循环迭代方式,快速收敛出全局最优学习率,进而得到最优组的权重信息,最终提高小目标车辆检测精度.实验结果表明,基于结构优化、超参数优化的YOLOv7-ECA-SSA检测模型在BDD100K数据集上的检测精度为79.01%,比原始模型提高了5.38%,具备更好的小目标车辆检测性能.
Abstract
To solve the small target under complex background of vehicle detection error detection,leak phenomenon,such as innovative put forward an improved YOLOv7 network target detection algorithm.Firstly,in order to solve the problem of small target vehicle secondary information interference,the ECA attention mechanism was integrated into the main network feature layer of YOLOv7 model,and the weight proportion of target area information was enhanced and irrelevant information was suppressed through adaptive learning.Secondly,in order to solve the qualitative problem of hyperparameter stochastic experience of neural network detection model training,the sparrow search algorithm was used to optimize the hyperparameter of detection model training,and the global optimal learning rate was quickly converged through internal and external double loop iteration,and then the weight information of the optimal group was obtained,and finally the detection accuracy of small target vehicles was improved.The experimental results show that the detection accuracy of YOLOv7-ECA-SSA detection model based on structure optimization and hyperparameter optimization is 79.01%on BDD100K data set,5.38%higher than that of the original model,and has better detection performance of small target vehicles.
关键词
车辆目标检测/YOLOv7/注意力机制/超参数优化/麻雀搜索算法Key words
vehicle target detection/YOLOv7/attention mechanism/hyperparameter optimization/sparrow search algo-rithm引用本文复制引用
基金项目
陕西省科技厅项目(2023-YBGY-031)
出版年
2024
NETL
NSTL
万方数据