改进Faster R-CNN模型的汽车喷油器阀座瑕疵检测算法

Automobile injector seat detection algorithm based on improved Faster R-CNN model

为完成喷油器阀座常见的瑕疵识别,对深度检测模型进行研究,提出基于Faster R-CNN模型的喷油器阀座瑕疵识别改进方法.首先,对常规生产下的喷油器阀座瑕疵图像进行采集、处理,构造出相关数据集;其次,在Faster R-CNN模型上对候选框和特征网络进行改进,获得比原有模型更高的精确度.实验结果表明:改进的Faster R-CNN模型在喷油器阀座瑕疵识别中精确度得到加强,识别精确度可达71.79%,相比原有模型精确度提升了近3.9%.说明该深度学习方法能够有效实现喷油器阀座瑕疵的识别,为后续自动一体化检测研究提供了基础.

The car injector seat plays a very important role in controlling the fuel quantity of the car.However,the detection of the injector seat is still carried out manually.The paper is to study the depth detection model to complete the common defect recognition of the injector seat.Due to the relatively small defect of injector seat,it is omitted in identification by the general detection model,which cannot meet the requirements of industrial production.Therefore,an improved method of defect identification of injector seat based on Faster R-CNN model is proposed.Firstly,the image of injector valve seat defect in conventional production is collected and processed,and the related data sets are constructed.Secondly,the candidate frame and feature network are improved in Faster R-CNN model to obtain higher accuracy than the original model.The experimental results show that the improved Faster R-CNN model is more accurate in identifying the defects of the injector seat,and the recognition accuracy can reach 71.79%,which is nearly 3.9%higher than that of the original model.This deep learning method can effectively realize the recognition of the defects of the injector seat,and provides a basis for the follow-up research of automatic integrated detection.