Research On Iris Location Based On Target Detection

The twenty-first Century is an era of rapid development of science and technology and data explosion.At the same time,the rapid development of modern science and technology also promotes the arrival of the era of Internet Science and technology.People are faced with a lot of data information every day,which has been integrated into everyone's life.However,a large part of these information needs to be highly valued by people,especially some bio information data related to personal information,compared with passwords and ID cards,is a unique sign of each person,a symbol of personal identity,and at the same time,compared with traditional password technology,bio information technology has higher stability and uniqueness.Therefore,for example,biological information data need to be treated with caution.Among them,iris recognition technology is widely concerned because of its unique technology.This also makes researchers directly face a problem that is the accurate extraction and recognition of personal information.We should adhere to the principle of high accuracy and high security in the treatment of biological information recognition.In the whole process of iris recognition system,the location algorithm of iris occupies a crucial position.Only by locating the inner and outer circles of iris accurately and quickly can we get high accuracy results in the subsequent feature extraction and recognition process.The iris location part is the most time-consuming part of the whole iris recognition system.The traditional location algorithm can only process a single image Because of the huge and complex computation,the positioning time is slow.In this paper,the concept of target detection based on deep learning is introduced into iris location system.According to the high-speed operation ability of neural network,not only the processing time of single image can be greatly reduced,but also the batch processing operation can be carried out,which further reduces the time of iris system positioning the overall image.With thedevelopment of deep learning technology in recent years,neural network has attracted extensive attention.The target detection algorithm also changes from the traditional manual feature extraction algorithm to the detection technology based on the depth neural network model.The main purpose of the target detection algorithm is to distinguish the parts we need from the other interference parts in the image,and then to pay attention to whether there are objects we need,if there are,label the monitoring box and mark the object category.Because the target detection algorithm has excellent performance and monitoring effect on the open source data set,this paper proposes an improved Yolo(you only look once)target detection algorithm based on deep learning to apply iris location.In this paper,Yolo model uses a convolutional neural network to directly predict different target categories and the location information of the target object,which has the characteristics of high speed and accuracy.Because the confidence degree of Yolo target detection algorithm's self evaluation standard is in a positive proportion to the intersection and parallel ratio of the target detection algorithm,it can well reflect the positioning performance effect of the network structure.So the kmeans algorithm in this paper uses a new calculation method to get the distance between the centroid box and the calibration box by calculating the intersection and union of the centroid box and the calibration box in the current training set instead of the Euclidean distance,and clusters the priori box with fixed length.On this basis,the classifier network predicts the prediction box.And according to the iris data set image must be large and clear,the basic size of fixed characteristics,this paper improved the network structure based on Yolo network,proposed a single scale network structure to train the image.At the same time,NMS,the non maximum suppression algorithm of the original model,is improved,and the hard threshold standard is replaced by the soft threshold standard,which makes the confidence of the overlapping detection frame attenuate,so as to avoid the problem of missing detection.Moreover,the average accuracy of this model is almost the same as that of the original Yolo model,but the positioning time is greatly improved.Compared with the traditional iris recognition and location algorithm,the iris location time of this system is greatly reduced,with real-time location ability and batch processing ability,which can be better applied to the iris real-time location system.Compared with Yolo model based ontransfer learning and SSD model,the improved Yolo model can be applied to iris recognition system(map)up to 98%.