In the automatic license plate recognition system, after the license plate is positioned and the characters are segmented, the segmented character image is identified. At present, common character methods mainly include character recognition methods based on template matching, character recognition methods based on neural networks, and character recognition methods based on statistical learning. The character recognition based on template matching determines the final recognition result according to the similarity between the character and the template. The algorithm is simple and mature, but its adaptation is not strong, and it is easy to cause misjudgment for the characters with breakage, pollution, adhesion and the like. The neural network based character recognition has good fault tolerance, self-adaptation and learning ability, but the sample training has a slower convergence speed. The character recognition method based on statistical learning extracts the characteristics of characters, and the classifier is designed to classify and recognize. The commonly used classifiers have SVM0, feature matching and other methods, which have higher recognition rate, but the recognition time is longer. This paper proposes an improved quadratic classification function (MQDF) classifier based on statistical model for license plate character recognition algorithm. The method is easy to design and implement, has good robustness and high recognition accuracy, and has short recognition time.
1 MQDF based on license plate character recognition 1.1MQDF classifier herein take research to improve the classification of the quadratic function (Modified recognition, machine learning, etc. Corresponding author: Liu Hui.
Classifier. The principle is as follows: the prior probabilities of each category are the same, and the Lyapunov central limit theorem is satisfied, so the probability distributions of various samples can be approximated by Gaussian distribution, then the quadratic discriminant function can be derived according to the Bayesian criterion. Excellent classifier. MQDF3 is proposed by Kimura et al. on the basis of QDF. KL transform is performed on QDF, and constants are used instead of small eigenvalues ​​to improve the calculation speed and the correctness of classification.
In equation (1), M is the number of categories, PU-) is the prior probability of the class, ((1) is the probability density function of the class, (is the joint density function, and P (independent categories; 3 1, D) is the eigenvector (descending order), 3 is the standard orthogonal matrix (33f = for small eigenvalues ​​with constant 5, instead, MQDF is: 7 in equation (7), k is the main eigenvector The above uses the Euclidean distance invariance: parameter 5, which is the average of the eigenvalues ​​mentioned by Kimura et al.
In a limited sample set, training QDF classifiers often affects the eigenvalues ​​of the class due to a variety of unstable noises, affecting the robustness of the classifier. Therefore, in MQDF, by replacing small eigenvalues ​​with constants, not only the classification efficiency is improved, but also the classification speed is increased and the memory used by the parameters is also reduced.
1.2 MQDF-based license plate character recognition Firstly, a sample library of each character is established, and then all the samples are trained by the MQDF method to obtain feature files of various characters.
After the pre-processed and character-divided image of the license plate to be identified, the normalized characters of the single character are sent to the MQDF classifier, and the classification result is obtained by performing the operation through the MQDF. Finally, the recognition result of each sub-classifier can be combined to obtain the recognition result of the entire license plate.
1.2.1 When the license plate character normalization training sample = 55, the recognition rate decreases because the eigenvalues ​​are arranged in descending order, and the subsequent eigenvalues ​​and eigenvectors are unstable noises; Simulation processing 3 CNC lathe test processing The verified NC program is transferred to the CNC lathe through the CAXA CNC car software to prepare the processing entity. Open AX-ADNC2011DNC, select "Send File", and after finding the NC program, click "Open". At this time, the CNC lathe is received through the serial port. After receiving, the tool is processed and machined on the CNC lathe. The result is shown in the NC program verification result. 4 Conclusion CAXA CNC is a CNC lathe machining programming and 2D graphic design software developed on a new CNC machining platform. . CAXA CNC car has powerful drawing function of CAD software and perfect external data interface. It can draw arbitrarily complex graphics. It can exchange data with other systems through DXF, GES and other data interfaces. H.CAXA CNC car has track generation and general post processing. Features. However, it is not easy to judge the interference of overcutting and knives generated during the processing of complex parts. The VERICUT software is used to simulate the NC program generated by the CAXA CNC car, which can effectively avoid overcutting and undercutting before machining, tools and workpieces, and machine tools. The phenomenon of fixture collision and other phenomena is of great significance for reducing production costs and improving product quality.
Leather Casual Shoes,Leather Sport Shoes Mens,Fashion Sport Shoes Leather,Mens Casual Leather Sneakers
GUANGZHOU ANAX FASHION SHOES LIMITED , https://www.anaxleathershoes.com