Research On Adaptive QPSO And Its Application In Image Classification

With the development of multimedia technology and computer network, massive digital images appear on the Internet, and they are closely related to our daily life. Researches on how to organize, manage and utilize those rich image resources have important theoretical value and practical significance for future digital service. The goal of image classification is to categorize a large number of digital images into a certain class automatically. Integrating other subjects'advanced theories into image classification, like machine learning, artificial intelligence, etc., to find an efficient method for image classification is still a crucial issue in computer vision, pattern recognition and other research areas.Based on the studies on Quantum-behaved particle swarm optimization (QPSO) and current situation of image classification, the contributions of this paper are as follows:First, this paper proposes an adaptive Quantum-behaved particle swarm optimization (AQPSO) based on QPSO. It evaluates the diversity of the swarm and changes evolutionary operators adaptively. It can prevent premature convergence to some extent and outperforms Quantum-behaved particle swarm optimization on standard functions.Second, this paper proposes a hybrid image classification algorithm based on AQPSO and support vector machine (SVM), which introduces AQPSO to optimize feature subset and parameter estimation for SVM at the same time, aiming to solving the problem of optimizing feature subset and parameter estimation simultaneously while training SVM classifier. The research's objective is to select the minimal feature subset without degrading the performance of SVM. Then, the paper tests the hybrid algorithm on real world image datasets. The experiment results show the hybrid algorithm proposed by this paper performs well and is effective.