为提高蒙皮损伤检测的自动化程度,提出一种基于改进YOLOv7通道冗余的机器视觉检测方法.首先针对飞机蒙皮损伤数据集背景单一的特点,提出增强型颈部特征融合改进算法,提高了飞机蒙皮损伤的识别精度和检测速度;其次针对主干特征提取网络的卷积通道冗余的问题,引入部分卷积PConv(Partial convolution),提出主干特征提取网络轻量化,减少模型的参数量,同时提高损伤的识别效率.试验部分首先在飞机蒙皮损伤数据集上探索了不同增强型颈部特征融合改进算法,确定了最优的改进方案;接着在飞机蒙皮损伤数据集上做消融和对比试验,改进算法与原YOLOv7算法比较,mAP(Mean average precision)提升了2.3%,FPS(Frames per second)提升了22.1 f/s,模型参数量降低了34.13%;最后将改进的YOLOv7模型与主流目标检测模型对比,证明了改进算法的先进性.
Detection for Aircraft Skin Damage Based on Improvement of Channel Redundancy for YOLOv7
In order to automatically realize detection of aircraft skin damage,a machine vision detection method based on the improvement of channel redundancy for YOLOv7 is proposed.Firstly,aiming at the characteristics of the single background for the aircraft skin damage dataset,an improved algorithm of enhanced neck feature fusion is proposed,which improves the recognition accuracy and detection speed of aircraft skin damage.Secondly,in order to solve the problem of convolution channel redundancy for the backbone feature extraction network,PConv(Partial convolution)is introduced,and the lightweight of the backbone feature extraction network is proposed,which reduces the number of parameters for the model and improves the efficiency of damage identification.In the experimental part,different improved algorithms of enhanced neck feature fusion were first explored on the aircraft skin damage dataset,and the optimal improvement method was determined.Then,ablation and comparative experiments were carried out on the aircraft skin damage dataset,and compared with the original YOLOv7 algorithm,the mAP(Mean average precision)is increased by 2.3%,the FPS(Frames per second)is increased by 22.1 f/s,and the number of model parameters is decreased by 34.13%.Finally,the improved YOLOv7 model is compared with the mainstream object detection model,which proves the advanced nature of the improved algorithm.
万方数据
