Nowadays,our society is in the era of big data.Massive data is generated in various domains and applications in every second,which contains valuable knowledge.However,such knowledge is unstructured and highly redundant,which makes it hard to directly acquire the underlying knowledge.Therefore,it is crucial to design an efficient and universal method for data analysis.Classification is one of the most import methods among available data analysis tools.It not only an important algorithm to recognize and categorize entities in the real world,but also an efficient way to classify and organize massive data.Support Vector Machine is a popular classification method which is widely used in various applications.Specifically,benefitted from its relatively complete theoretical support and good classification performance in data set,Support Vector Machine had been applied extensively in a lot of profession.However,the hyper-parameter selection has a large impact on the classification accuracy and usually restricts the generalization ability of the model.Recently,it has become a hot topic to improve the classification accuracy and generalization ability by improving the hyperparameter setting.Particle Swarm Optimization algorithm is a swarm intelligence evolutionary algorithm based on the foraging behavior of birds.When we use this algorithm to search,each particle among groups can searches independently and changes information between each other currently.The parallelism makes the algorithm have the advantages of fast search speed and high precision,which has attracted the attention of many scholars.It has been empirically observed that Particle Swarm Optimization algorithm has some advantages in parameter search.However,there are some problems in the searching process of basic Particle Swarm Optimization algorithm,such as easily falling into local optimum and reducing population diversity.This graduation paper improved the basic Particle Swarm Optimization algorithm and applied it to the parameter search of Support Vector Machine.Finally,the improved algorithm was applied to the actual industry project of film bubble detection on the touch screen of mobile phone.(1)A Particle Swarm Optimization algorithm improved by genetic immunity(GAIPSO)is proposed.It avoids falling into local optimum and reduces population diversity in the optimization process of basic Particle Swarm Optimization algorithm.The algorithm is improved in the following three aspects: 1)It introduces genetic crossover with immune mechanism into basic Particle Swarm Optimization algorithm to improve the population diversity reduction due to group influence.2)It enhances neighborhood search ability and improving breakaway ability when particles fall into local optimum.3)It sets up particle crossing boundary processing mechanism to eliminate the adverse effects of particles beyond the boundary.This study uses the benchmark function to the GAIPSO algorithm simulation experiment,and the superiority of GAIPSO algorithm in the optimization accuracy and convergence speed is verified.(2)The proposed algorithm solves the problem that the choice of SVM parameters affects the classification accuracy and generalization ability of the model.Firstly,this study used the combination kernel function to replace the single kernel function.This technology allows it to search more suitable mapping space.Then,the GAIPSO algorithm is applied to search the parameters of SVM algorithm.Tested by UCI manual data set,the results showed that the classification accuracy and generalization ability of SVM algorithm was improved after the parameters optimized by GAIPSO algorithm.(3)The proposed algorithm solves the problem that large number of false detection exists in industrial defect object detection.It replaces the softmax function of frequently-used depth object detection method Faster-RCNN by the Support Vector Machine optimized by GAIPSO algorithm.Finally,it is used in the actual industry project of film bubble detection on the touch screen of mobile phone,and the detection result showed that the improved detection method enhances the detection accuracy and has practical value for the actual production with the decrease of false positive rate,filtration rate and the rise of rolled throughput yield. |