| As a very important information medium in the field of manufacturing and logistics,barcode plays a very important role in our daily life.The traditional barcode identification method usually requires workers to hold a scanner to identify barcodes,which has high reliability but low efficiency and cannot identify multiple barcodes at the same time,while in the actual industrial and logistics fields,there are often situations where multiple barcodes need to be read at high speed at the same time.Therefore,it is of high theoretical research and practical application value to design an algorithm that can quickly locate and recognize barcodes in complex environments.This paper takes the barcode detection task based on image processing and computer vision technology as the research content,aims at improving the accuracy of barcode detection and real-time of algorithm in complex background,studies and analyzes the task difficulties of barcode detection,and combines traditional image processing technology and deep learning computer vision technology to perform fast barcode detection and recognition on the images of goods collected on the fast sorting line.The main research contents of this paper are as follows.1.Established a barcode detection database that conforms to the characteristics of the logistics sorting line.Since deep learning algorithms require a large amount of labeled data for training,it is necessary to establish a highquality barcode detection database.In this paper,a large number of commodity barcode images are collected by industrial cameras on the logistics sorting line.After data cleaning,the barcode area is accurately marked by professional data labeling software,and a high-quality barcode detection database is obtained.2.Two types of barcode positioning algorithms are designed to meet the needs of different occasions.Barcode positioning is the first step in barcode detection,the purpose is to accurately locate the barcode area in the image.In this paper,two types of barcode positioning algorithms are designed based on target detection algorithm and image segmentation algorithm,which can be applied to different detection occasions.The barcode positioning algorithm based on target detection is a two-stage algorithm with high detection efficiency but poor robustness,and is suitable for occasions that require high detection speed.Imagebased image segmentation is an end-to-end method that uses a neural network to segment the barcode area in the image,which has high precision but a large amount of calculation,and is suitable for occasions with high detection accuracy.3.Design a lightweight neural network training strategy based on structural similarity constraints.In order to improve the inference speed of the barcode location algorithm based on the image segmentation method and realize fast barcode detection,this paper uses the knowledge distillation algorithm to train the lightweight image segmentation network.In this paper,the loss function is designed to maintain the structural information and texture information of image features,and the image feature information is transferred from the teacher network to the lightweight student network,which enables the student network to reduce the number of parameters while still maintaining a strong image segmentation ability.4.Design a barcode decoding algorithm based on feature vector matching.In order to overcome the shortcomings of the traditional decoding method that is sensitive to the width error of the black and white stripes of the barcode,this paper proposes a decoding method based on feature vector matching.First,the binary information of the barcode is extracted through image denoising and binarization operations,and the binary sequence is converted into Feature vectors and vector matching based on similarity,thereby improving the accuracy of decoding results. |
