Research On Rapid Detection Method Of Micro-target Ball Based On Visual Entropy Attention Mechanism

With the development of machine vision technology,the requirement of detection accuracy and automation in the field of micro-vision is becoming higher and higher.The disadvantage of traditional manual precise detection is that the error is large and the efficiency is low.Machine vision technology,with its unique characteristics of non-contact,high precision and so on,has gradually occupied more and more markets;and its application in the industrial field has gradually increased.(1)Firstly,the method of multi-field cooperative detection is adopted,and the large field image of the micro-target ball is acquired by using a lens with relatively low resolution.Then,the visual attention algorithm is applied to acquire the region of interest in the large field of view image,and then the coordinates of the region of interest are obtained to gain a small field of view.Secondly,the features of micro-sphere are extracted in the small field of view,and finally integrated into the large field of view to obtain the micro-target spherical coordinate position.(2)In the research process of visual attention algorithm,this paper takes the micro-visual environment as the research background,and combines the theory of information theory with machine vision based on the principle of precision and efficiency,and proposes a visual attention algorithm based on entropy calculation method.The derivation of algorithm is based on the residual light perception mechanism of human-eye in biology.Combining the theory of probability theory with the entropy value of information theory,the visual entropy attention algorithm is obtained,and the algorithm is used to lock the region of interest in the large field of view.The experimental results show that in the micro-visual environment,the visual entropy attention mechanism can accurately and efficiently acquire the salient regions in the large field of view.(3)Due to the low signal-to-noise ratio of micro-visual images,it is necessary to perform image enhancement processing to obtain clearer and more accurate micro-sphere images.However,due to the high degree of aggregation of the micro-target spheres,most of the field of view has no distribution of micro-target spheres.In order to reduce the influence of noise in the target-less area on the overall image,this paper adopts the method of local image enhancement,which is a combination of target edge enhancement and illumination enhancement.Experiments in the paper show that the scheme applied in the aspect of image enhancement can effectively enhance the display effect of the micro-target sphere compared with other methods.(4)After successfully acquiring large and small fields of view,feature extraction and matching processing are performed on the targets of large and small field of view images.However,due to the large data volume and low SNR of micro-visual images,the traditional feature extraction matching algorithm cannot be accurately obtained.Therefore,on the basis of traditional algorithm,this paper proposes a target contour difference-entropy convergence and discriminant algorithm,which can eliminate the misidentification and mismatching and improve the accuracy.