Research Of Electronic Components Detection Algorithm Based On Robot Vision

Along with the fast-speed development of information technology and economy,the speed of electronic product replacement is enormously quick,and electronic components are becoming more and more miniaturized and integrated.Therefore,the difficulty of visual detection of electronic components is greatly improved.In recent years,object detection has undergone unprecedented progress,but the detection performance is still a huge gap between small and large objects due to the problems of small objects that small object images have many changes in scale,low resolution,blurred contour and lacking information.In this paper,by analyzing the difficulties of electronic components in visual object detection,aiming at the characteristics of small size of electronic components and taking the electronic components in production line as the research object,a color image motion fuzzy deblurring algorithm based on HSV color space and a feature deconvolutional-fuse and feature enhancement improvement SSD algorithm,named FDESSD,are proposed to detect electronic components,which is verified to meet the requirements on the simple hardware platform.The main research contents are as follows:1.Motion blur deblurring algorithm for color images based on HSV color space is proposed.In view of the problem of motion blur in the production line motion or camera motion,motion fuzzy removal algorithm is studied.Because the effect of motion blur on each channel in HSV color space is different,a motion fuzzy deblurring algorithm for color images based on HSV color space is proposed.In different channels of HSV color space,Radon transform estimator based on spectrum edge detection is used to obtain fuzzy kernel,and then the deblurring restoration based on Richardson-Lucy guided filter is carried out in each channel.2.Feature Deconvolutional-fuse and Feature Enhancement improvement SSD algorithm is proposed.In view of the small size of electronic components and the poor performance of the state-of-art object detection algorithm in small objects,the shortcomings of SSD algorithm are analyzed in detail,and Feature Deconvolutional-fuse and Feature Enhancement improvement SSD algorithm,named FDESSD algorithm,is proposed by fusing the high layer including more semantic features and the low layer and using SE(Squeeze and Excitation)module to enhancement features to improve the detection effect of the small target.Through experiments,different modules are analyzed,and the performance of the algorithm is evaluated by using the MSCOCO dataset.Through training and testing,FDESSD achieves 50.8% m AP(Mean Average Precision),7.7% higher than SSD,which is proved that the FDESSD has improved the detection effect of small object.3.Evaluation experiment of detection system for electronic components is carried.In response to the problem that there is no public available dataset for electronic component detection at present,a simple experimental environment is built using industrial camera,lens,lighting scheme and PC and the Electronic Component Dataset(ECD)is established by selecting 17 common electronic components.Then the image quantity rationality of the dataset is evaluated,and the uniformity of the dataset is proved by using 4-fold cross-validation.The FDESSD algorithm is superior to state-of-art object detection algorithm on this dataset.