Research And Application Of Small Object Detection Method For Workpiece Defects

Workpiece defect detection refers to the inspection of workpiece surface quality in the process of machining.It is an important part of ensuring product quality in the field of industrial manufacturing.At present,manual detection is mainly used in workpiece defect detection,which has some problems,such as inconsistent standards,high cost,low efficiency and fatigue of detection personnel.In addition,the defect detection method based on traditional image processing technology also has a certain degree of application.This method has high efficiency and can achieve good results when the discrimination between the defect and the background is high.However,when the image background is complex,the interference is large,the defect area is small and not obvious,the detection effect is not good.Deep learning can explore the rules contained in the image,identify the specific patterns of defects,and provide a new method for defect detection.Aiming at the problem of poor detection effect of weak small target in workpiece defect detection,this paper applies object detection technology based on deep learning to workpiece defect detection,and proposes two defect detection methods.The first is a small defect detection method based on feature pyramid.In order to fully retain the information of small defects in the image,firstly,several basic convolution blocks are used to extract the defect features to form multi-level defect features.In order to fully obtain the defect context information,bidirectional feature pyramid is introduced to integrate different levels of defect information,and a workpiece defect detection network structure is constructed.In the model training stage,the loss feedback mechanism is used to solve the problem of sample imbalance.When the loss ratio caused by small defects is less than the threshold value,the input part stitches the images,and stitches the four images with half of the length and half of the width into one image of the original normal size,and then inputs it into the network for training.The second is the detection method of hidden small defects.Hidden small defects refer to the defects with complex image background,large interference,small defect area and not obvious enough.Aiming at the detection problem of this kind of defects,the defect detection is divided into two modules,the first is to search defect features and carry out rough positioning,and the second is to carry out accurate positioning.The defect feature part is extracted to simulate the animal vision system,and a multi branch receptive field module with different sizes is designed,which can accurately find the defects in the image.In the precise positioning stage,attention mechanism and decoder are used to locate the defects accurately.Self made image set of automobile engine cover.The front and back images of the cover are collected by using the area array camera and the ring light.When the side images are collected,the cover is rotated and photographed by using the linear array camera.After obtaining the original image,the cover image set is made by preprocessing.The two methods proposed in this paper are tested on the front and back image sets and the side image sets respectively,and the results show that the proposed method achieves very good results in accuracy and efficiency.Based on the two kinds of defect detection models,the workpiece defect detection system is developed,which mainly realizes the functions of equipment condition monitoring,detection result statistics,real-time display of detection effect,and can basically meet the needs of workpiece defect detection.