Research On Image Detection Algorithm Of Industrial Bar Code Based On SSD Algorithm

With the rapid development of the Internet and the continuous expansion of the Internet application scenarios,bar code as a carrier of information,it’s use frequency and application areas are also expanding.At present,bar code technology has become an indispensable technology in human production and life.As the largest manufacturing country in the world,bar code is widely used in the industrial field,which provides convenience for industrial information integration,but also brings the high efficiency and accuracy requirements of barcode recognition technology.In actual industrial production,natural lighting is poor,and have many active lighting sources,vibration and other influencing factors,resulting in barcode positioning failure.Aiming at the practical application requirements of industrial bar code detection,this thesis carried out a series of researches on how to improve the model detection performance based on SSD model.The main work and innovations of this thesis are as follows:(1)Industrial bar code detection equipment has very high real-time requirements for model detection.Based on SSD model,this thesis constructed a lightweight SSD model by reducing the number of model channels to improve the detection efficiency of the model.And according to the characteristics of industrial bar code image,set up a more consistent with the data prior frame.Meanwhile,the positioning loss function is optimized in order to improve the detection accuracy.(2)Due to the sharp decrease in the number of model channels,the detection accuracy decreases greatly.However,SSD model adopts multiscale feature prediction,and its single-layer detection layer has single feature,which leads to weak effective information of feature map and serious missed detection of difficult samples.An intensive multi-scale feature fusion method is proposed in this thesis.The shallow detection layer has both the geometric information representation ability of shallow features and the strong semantic information of deep features to improve the performance of model detection.(3)In order to improve the detection accuracy of industrial bar codes,the attentional mechanism is embedded based on intensive multi-scale feature fusion.The channel attention mechanism SENet and the mixed domain attention mechanism CBAM act on each detection layer after feature fusion respectively to locate and strengthen the interested information and suppress useless information.In this thesis,several experimental results show that the improved network model has certain advantages in detection accuracy and real-time,which verifies the feasibility and superiority of the model constructed in this thesis.